Want to work in Artificial Intelligence? Don’t fancy the grind of the commute and awkward water cooler conversations? Good news, there are plenty of AI roles out there which you can do from home, with big tech companies.

They pay well too, providing you’ve got the resume to justify being appointed to one of these roles, with the likes of Google, Microsoft and Nvidia paying top dollar to attract AI expertise and talent.

While remote work policies vary from company to company, there are still plenty of flexible employers looking for workers. Some aren’t quite as flexible as you might like, and we’ll point them out to you. Good luck with the search!

Remote AI Jobs at Microsoft

Microsoft, as you may expect, has plenty of AI jobs to offer. The company is currently making a massive push for AI across its product range, and is heavily invested in the technology. 

At the time of writing, there are over two thousand roles globally at the company with an AI focus, with a wide range of salaries, so there are plenty of options. Of those, just under three hundred are fully remote.

We’ve picked a few out that caught our eye, but for the full rundown, take a look at the dedicated Microsoft job site and filter by AI/0 days in office.

Principal Software Engineer – AI Infrastructure

• Base Salary Range: $165,600 – $296,400 / year
• Focus: Large-scale distributed runtime infrastructure for foundation model training, multi-agent orchestration, and GPU compute efficiency.

Senior Software Engineer – AI Frameworks

• Base Salary Range: $165,600 – $331,200 / year
• Focus: Optimizing, scaling, and maintaining deep learning compilation frameworks and open-source models in collaboration with research teams.

Senior Software Engineer – LLM Serving & Performance

• Base Salary Range: $119,800 – $261,000 / year
• Focus: Low-level kernel optimization, inference runtime latency reduction, and memory layout strategies for real-time model serving.

Software Engineer II – Copilot & Core AI

• Base Salary Range: $98,000 – $186,000 / year
• Focus: Integrating generative AI endpoints into Microsoft 365, enterprise agent orchestration platforms, and cloud microservices.

Principal Product Manager – Generative AI

• Base Salary Range: $158,000 – $282,000 / year
• Focus: Defining product architecture, evaluation benchmarks, and safety controls for next-generation Copilot tools and enterprise agent workflows.

AI & Data Security Specialist

• Base Salary Range: $112,000 – $218,000 / year
• Focus: Designing zero-trust governance frameworks, red-teaming protocols, and privacy-preserving data layers for enterprise LLMs.

Senior Data Scientist – Microsoft AI (MAI)

• Base Salary Range: $125,000 – $240,000 / year
• Focus: Post-training alignment, reinforcement learning from human feedback (RLHF), and evaluation metrics across large language models.

Solution Engineer – Azure AI Solutions

• Base Salary Range: $110,000 – $205,000 / year
• Focus: Direct enterprise client architecture for custom Copilots, Azure AI Foundry workloads, and high-performance inference pipelines.

Training Manager – AI Experiences

• Base Salary Range: $105,000 – $210,000 / year
• Focus: Developing organizational AI fluency programs, compliance training, and adoption frameworks for enterprise platforms.

HR Business Partner – Microsoft AI Division

• Base Salary Range: $106,400 – $222,600 / year
• Focus: Talent strategy, organizational planning, and human capital growth for Microsoft’s dedicated AI business unit.

Remote AI Jobs at Google

If you’re looking for a fully remote Google role, we’ve got bad news – they aren’t easy to come by. Google is one of the companies that fully embraced the post-pandemic return to office movement, and as such, work from home jobs are rare. 

Most Google roles demand that employees are in the office three days a week, but workers are granted two days remote each week. We’ve picked out some jobs below that are remote eligible. That doesn’t mean that these will be guaranteed fully remote, but rather that Google is willing to make exceptions to their in-office policy.

What this means will depend on the department, the hiring manager, and other factors, but it’s worth shooting your shot if you want to work for the big G. Check out the Google careers portal.

1. Staff Software Engineer, Agent Development Lifecycle (Google Cloud AI)
   • Focus: Architecting end-to-end orchestration pipelines for autonomous agents, tool integration, and LLM evaluation within Google Cloud Platform.
   • Base Salary Range: $207,000 – $301,000 / year + bonus + RSUs
2. AI Engineer, Advanced Solutions Lab (Google Cloud)
   • Focus: Partnering with enterprise customers to build, fine-tune, and deploy custom agentic solutions and generative AI models on Vertex AI.
   • Base Salary Range: $138,000 – $210,000 / year + bonus + RSUs
3. Senior Software Engineer, AI/ML (Google DeepMind / Core AI)
   • Focus: Developing low-level distributed runtime frameworks, model fine-tuning tools, and deployment pipelines for Gemini foundation models.
   • Base Salary Range: $161,000 – $239,000 / year + bonus + RSUs
4. Software Engineering Manager, SensAI (Sensors & AI)
   • Focus: Leading engineering teams building sensor fusion algorithms, heuristic ML, and spatial computing infrastructure.
   • Base Salary Range: $220,000 – $315,000 / year + bonus + RSUs
5. Research Scientist, Generative AI & Multimodal Models
   • Focus: Advancing core architecture design for multimodal LLMs, RLHF alignment methods, and next-generation reasoning capabilities.
   • Base Salary Range: $180,000 – $285,000 / year + bonus + RSUs
6. Technical Program Manager, Cloud AI & Foundation Models
   • Focus: Driving cross-functional release schedules for Gemini API integration, GPU/TPU cluster allocation, and safety evaluations across teams.
   • Base Salary Range: $138,000 – $198,000 / year + bonus + RSUs
7. Customer Solutions Engineer, AI & Data Analytics
   • Focus: Designing technical proof-of-concepts, custom LLM pipelines, and AI cloud architecture for strategic Google Cloud enterprise accounts.
   • Base Salary Range: $124,000 – $186,000 / year + bonus + RSUs
8. Product Manager, Vertex AI & Developer Tools
   • Focus: Managing product roadmaps for developer APIs, fine-tuning environments, and model governance toolkits on Google Cloud.
   • Base Salary Range: $155,000 – $248,000 / year + bonus + RSUs
9. Security Engineer, Generative AI Safety & Red Teaming
   • Focus: Conducting threat modeling, jailbreak vulnerability assessments, and automated red-teaming for foundational models prior to deployment.
   • Base Salary Range: $145,000 – $225,000 / year + bonus + RSUs
10. Data Scientist, AI Quality & Benchmarking
    • Focus: Designing quantitative metrics, human-in-the-loop evaluation frameworks, and model performance baselines across generative AI products.
    • Base Salary Range: $130,000 – $205,000 / year + bonus + RSUs

Remote AI Jobs at Nvidia

Nvidia is fuelling the AI revolution through its chips, so if you want to be at the forefront of the AI industry, this is a good place to make a name for yourself.

Nvidia’s approach to remote working is fairly flexible so you won’t have any issues finding a fully remote job. Getting it however, will be a different matter, as competition is fierce. We’re not trying to put you off, but the company receives 10 million applications a year, and only accepts 10,000 of them.

If you’ve got what it takes, check out the roles below, or head to the dedicated vacancies portal.

Senior Software Engineer – AI Runtimes & Inference (TensorRT / Triton)

• Focus: Optimizing low-level LLM execution, kernel compilation, and dynamic batching inside TensorRT-LLM and Triton Inference Server.
• Base Salary Range: $184,000 – $287,500 / year + RSUs + bonus

Principal AI Infrastructure Engineer – DGX Cloud

• Focus: Building distributed cloud platforms, cluster orchestration engines, and high-throughput communication layers for large-scale GPU clusters.
• Base Salary Range: $224,000 – $356,500 / year + RSUs + bonus

Senior Research Scientist – Foundation Models & Generative AI

• Focus: Fundamental research in multi-modal LLM architecture, synthetic dataset curation, post-training alignment (RLHF/DPO), and world models.
• Base Salary Range: $180,000 – $300,000 / year + RSUs + bonus

Senior Solutions Architect – Generative AI & Enterprise LLMs

• Focus: Partnering with enterprise clients to design production-grade deployments using Nvidia NeMo Framework, NIM microservices, and Retrieval-Augmented Generation (RAG).
• Base Salary Range: $148,000 – $230,000 / year + RSUs + bonus

Senior Deep Learning Software Engineer – Autonomous Systems / Robotics (Isaac AI)

• Focus: Developing computer vision, spatial AI, and physics-based simulation algorithms powering Nvidia Isaac Sim and Omniverse.
• Base Salary Range: $170,000 – $276,000 / year + RSUs + bonus

Technical Product Manager – NeMo & AI Microservices (NIM)

• Focus: Defining developer API roadmaps, enterprise model guardrails, and microservice architectures for deploying foundation models on containerized infrastructure.
• Base Salary Range: $152,000 – $241,500 / year + RSUs + bonus

Senior Developer Relations Engineer – GenAI & LLM Ecosystem

• Focus: Working directly with open-source AI maintainers, startups, and developers to accelerate model training and fine-tuning on Nvidia hardware.
• Base Salary Range: $144,000 – $224,250 / year + RSUs + bonus

AI Safety & Alignment Engineer – Guardrails & Red Teaming

• Focus: Designing automated testing pipelines, jailbreak defense systems, and input/output filtering frameworks (NeMo Guardrails) for enterprise deployments.
• Base Salary Range: $160,000 – $250,000 / year + RSUs + bonus

Data Scientist – AI Model Evaluation & Performance Benchmarking

• Focus: Creating standardized evaluation suites, measuring token latency/throughput across hardware setups, and benchmarking foundation model quality.
• Base Salary Range: $138,000 – $218,500 / year + RSUs + bonus

Technical Program Manager – Accelerated Computing & AI Software

• Focus: Managing cross-functional release pipelines across CUDA libraries, AI frameworks, and GPU cloud infrastructure teams.
• Base Salary Range: $135,000 – $210,000 / year + RSUs + bonus

Remote AI Jobs at Apple

Of the tech companies we’ve covered here, Apple is the strictest when it comes to working remotely. Simply put, it isn’t that flexible. You’ll find that most Apple roles demand employees are in the office three days a week, typically Tuesday to Thursday.

It’s been something of a contentious issue, no more so than in 2022, when Apple’s then Head of Machine Learning, Ian Goodfellow, threw in the towel with the company, citing its return to office policy. He said “I believe strongly that more flexibility would have been the best policy for my team.”

So there you go. If the Head of Machine Learning at Apple can’t catch a break and work fully remotely, there’s not much hope for anyone else. In view of this, we’ve listed some roles below that do offer two days a week remote. When it comes to Apple, that’s probably about as good as it’s going to get. See the full list at Apple Careers.

Machine Learning Engineer, Foundation Models (Apple Intelligence)

• Focus: Training, optimizing, and deploying on-device and server-based foundation models powering Apple Intelligence features across iOS and macOS.
• Base Salary Range: $170,000 – $295,000 / year + RSUs + bonus

Research Scientist, Generative AI & Multimodal LLMs

• Focus: Conducting fundamental research in vision-language models, reasoning capabilities, and post-training alignment (RLHF/DPO).
• Base Salary Range: $185,000 – $320,000 / year + RSUs + bonus

Computer Vision Engineer, On-Device Intelligence

• Focus: Building ultra-low-power, hardware-accelerated computer vision algorithms for visual search, scene understanding, and spatial computing.
• Base Salary Range: $160,000 – $275,000 / year + RSUs + bonus

Speech & Natural Language Processing (NLP) Engineer, Siri

• Focus: Enhancing conversational understanding, multi-turn dialogue management, and contextual personal intelligence for Siri.
• Base Salary Range: $155,000 – $265,000 / year + RSUs + bonus

Machine Learning Infrastructure & Compute Systems Engineer

• Focus: Building distributed training frameworks, GPU/TPU cluster management platforms, and model compilation toolchains for Apple Silicon.
• Base Salary Range: $175,000 – $300,000 / year + RSUs + bonus

Privacy-Preserving Machine Learning Engineer

• Focus: Developing Private Cloud Compute architecture, differential privacy mechanisms, and secure enclave processing for generative AI workflows.
• Base Salary Range: $165,000 – $285,000 / year + RSUs + bonus

Engineering Program Manager, Generative AI Platforms

• Focus: Managing cross-functional delivery schedules across research, core OS teams, hardware engineering, and privacy compliance.
• Base Salary Range: $145,000 – $240,000 / year + RSUs + bonus

Machine Learning Data Operations Lead

• Focus: Managing large-scale data pipelines, synthetic data generation, and human evaluation benchmarks for foundation model quality.
• Base Salary Range: $135,000 – $220,000 / year + RSUs + bonus

Applied Scientist, Personalization & Search Intelligence

• Focus: Designing intelligent ranking, recommendation, and indexing models across Spotlight, App Store, and Apple Music.
• Base Salary Range: $160,000 – $270,000 / year + RSUs + bonus

Product Manager, Apple Intelligence Experiences

• Focus: Defining user interaction paradigms, evaluation safety criteria, and feature roadmaps for system-wide AI capabilities.
• Base Salary Range: $165,000 – $275,000 / year + RSUs + bonus

Marketers — like everyone — are still waiting for AI to make a significant impact, with a new study finding that most professionals haven’t seen a big improvement just yet.

The study from Bain noted that only 6% of marketing professionals say that AI has influenced performance in a significantly positive way, despite 47% of leaders noting that it is a “core capability” for the industry.

So, how are these 6% of professionals making a marketing difference? They focus on providing top-down, centralized AI strategy, rather than just throwing tools at employees and hoping for the best.

AI Isn’t Moving the Needle for Marketers

According to the study from Bain, titled AI in Marketing: How Leaders Achieve Double the Revenue Impact, research found that only 6% of marketing professionals say AI is having a significant impact on performance.

This lack of significant performance improvements points to a disconnected strategy, not a lack of adoption. The study also found that 47% of marketing leaders say that AI is a core capability for businesses in the industry.

The report “surveyed 1,397 CMOs, chief financial officers, and senior marketing and finance executives across technology, consumer, retail, financial services, media, applications, education, landmark, and home consumer services.”

What Marketing Leaders Do Different

The study from Bain focused on marketing leaders — who are seeing 11%+ annual revenue growth — and marketing laggards — who record flat or declining revenue growth.

Marketing leaders had some very specific tendencies that separated them from laggards, generating AI results that positive impact their business.

They are 8.5 times more likely to experiment with AI, they are 1.5 times more likely to use AI to enhance customer personalization, and they are two times as likely to allocate more than 25% of budget to AI.

How Marketing Leaders Use AI

If you want to make sure that your business is in that 6%, we don’t blame you. After all, AI is expensive, and the last thing you want to do is waste resources on technology that isn’t moving the needle.

The study from Bain found that there are some specific steps that marketing leaders take to optimize their AI usage.

• Centralize AI strategy: Marketing leaders align all business priorities with AI, allowing the entire company to understand basic standards and protocols.
• Rebuild workflows and teams: Marketing leaders evaluate and reevaluate workflows in hopes of finding ways to improve them with AI.
• Focus on customer value: Marketing leaders know that the key to impact is produce customer value, so they align strategy with customers in mind.

Simply put, you can’t just inject AI into your business and pray that it has an impact. This technology is world change, which means you’ll need to change your world to get the most of it.

Check out our guide to AI training courses to learn more.

Maybe leave ChatGPT out of your enterprise purchasing decisions, because a new study found that these AI models are wrong 86% of the time when it comes to shopping.

Models like Gemini, Claude, Perplexity, and yes ChatGPT were all tested and were consistently found to provide incorrect prices and specs and suggest outdated products.

Even worse, the models weren’t just a little bit wrong; they really whiffed it. When asking about high-cost items like laptops, mattresses, and TVs, AI was incorrect by a median average of $300.

How Accurate Is AI for Shopping?

The study from Product.ai investigated the free and paid tiers for ChatGPT, Claude, Gemini, and Perplexity, asking each of them the same 220 shopping questions.

According to the report, the AI models provided conflicting and incorrect information in responses 86% of the time.  Even worse, comparison questions that specifically matchup products against each other were incorrect 97% of the time.

How Wrong Is AI for Shopping?

The study goes on to note that AI models aren’t just a little wrong when it comes to prices, features, and specs of high-end items like laptops, mattresses, and TVs. They were way off.

“When a price was wrong, it was wrong by a median of $300…There was almost no middle ground between exactly right and badly wrong.” – Product.ai researchers

On top of that, one out of every ten of these incorrect responses were wrong by as much as $500, which isn’t a small error; that’s a serious problem.

Why This Matters

Right now, this very second, there are people using AI to make purchases. One study from Bain & Company noted that around “30% to 45% of US consumers are using generative AI for product research and comparison.”

If that significant portion of shoppers are getting recommendations and pricing/feature information that is inaccurate, we could be talking about millions or even billions of dollars being misspent.

That doesn’t exactly create a positive customer experience. And in an era where trust is more valuable than anything for businesses, it might be better to leave AI alone for the time being.

Both OpenAI and Anthropic have this week revealed cheaper AI models, as competition heats up in the market, especially from Chinese models such as Alibaba and Deepseek.

The new models are claimed to be around 40-50% cheaper than previous versions, with both companies aiming to offer solutions that are more token efficient.

With the price of using AI on the rise for US businesses, and the threat of ‘Tokenmaxxing‘ burning through AI budgets, the move will be a welcome one for many.

OpenAI Launches GPT-6 Sol and GPT-6 Luna

Hot on the heels of its Astra model earlier in the month, OpenAI announced the addition of two new models, GPT-6 Sol and GPT-6 Luna, designed to be budget-friendly options.

The secret to the lower cost, according to OpenAI, is that these models are more efficient to run than previously, and as a result they’ve been able to set the price accordingly. As noted in the press release for the new models: “Improvements in caching and inference let us serve these models at lower cost, and we’re passing those savings directly on to users.”

According to OpenAI’s own benchmarking, GPT-6 Sol outperforms Claude Opus 5 at 9% of Opus 5’s cost per task. GPT-6 Luna goes even further, with an apparent 58% lower cost per task than its predecessor.

Factual error rates have also reportedly been improved. GPT-6 Sol makes ‘about half as many mistakes’ as its predecessor, which OpenAI puts a near Astra-level reliability, but at a much lower cost.

GPT‑6 Sol and GPT‑6 Luna are out now in ChatGPT Work and Codex, available to Plus, Pro, Business, Enterprise, and Edu users.

Anthropic Launches Claude Opus 5.5

Not to be outdone, Anthropic revealed its cheaper model on the very same day as the OpenAI announcement, with Claude Opus 5.

There are a range of improvements with the model, but undoubtedly the headline is that it’s 40% cheaper than its predecessor, marking a potentially massive saving for heavy users. Tokens cost less, and the model also requires few tokens per task.

In its own tests, Anthropic claim huge improvements over Opus 5. It reports that testers were able to build working games with just one prompt, and that another completed a 680,000-line code migration in a day, using the model. Another test saw Opus 5.5 tasked with reducing load times across a web app, and it was successful in 39 out of 40 instances, while the previous model made smaller improvements that ‘altered the apps behavior’.

Safety, something of a hot topic right now, has also improved, with Anthropic claiming the new model is much less likely to deviate from instruction, or make ‘hard to reverse’ choices.

Claude Opus 5.5 is available now.

Threat from Chinese Competition

The announcement of these low cost AI models from Anthropic and OpenAI seems to be a reaction to overseas competition, which is moving at a breakneck speed in a bid for AI dominance. China is at the forefront, with models such as Deepseek and Alibaba offering highly capable tools at a fraction of the cost of US models.

The rivalry between China and the US as both seek AI dominance is also generating safety concerns, with President Trump meeting China’s president Xi Jinping this week in Washington. The main topic of conversation is likely to be how to keep up the AI development pace safely without the need to slow down, something which many in the AI industry have called for.

The introduction of these lower cost models will be good news for businesses that are using AI heavily on a daily basis, but does it go far enough to stop users jumping ship and moving to even cheaper foreign models?

AI has improved productivity for a lot of individuals, but businesses haven’t seen the fruits of that labor yet. Why? Because employees keep lobbing AI slop grenades at each other.

What are AI slop grenades? The term has been gaining steam, with the CEO of Shopify mentioning it in an interview with CBS News, noting that it is “a bad thing.”

The term slop grenades refers to the all-too-common phenomenon that sees AI-generated work being completely unedited and unchecked before passing it on, leading to additional hours spent reviewing and correcting it.

AI Slop Grenade: Term Explained

An AI slop grenade is a term that describes AI-generated work completed by an employee that does not check, review, or edit the work before sending it on to other employees.

In essence, the AI-generated work — which we know can often be filled with errors — acts as a ticking time bomb, creating the appearance of accomplished work, but then exploding into a mess that requires additional work from other teams.

This phenomenon has been described in quite a few ways. Just a few months ago, many were describing it as the AI rework tax, explaining that any productivity benefits gained from using the technology were “taxed” by the additional work required to make it acceptable by normal business standards.

Shopify CEO: Slop Grenade ‘a Bad Thing’

The term is gaining notoriety because it was used by the CEO of Shopify in an interview with CBS News last week to describe the increasingly problematic trend he sees in the business world.

“You don’t really read it, and now it has to be reviewed by your colleagues, and they are like, ‘this doesn’t look right.’ You’re just letting AI do the work for you.” – Tobias Lütke, CEO of Shopify

With 40% of AI work often still requiring rework, it’s understandable that these slop grenades would have such a negative impact on productivity.

The Great AI Rollback

Just a few short years ago, Shopify was all in on AI. The CEO infamously told employees that they need to prove that they “cannot get what they want done using AI” before doing anything else.

Shopify isn’t the only company that has rolled back its AI commitment. Duolingo and Salesforce both infamously laid off a huge number of employees in favor of AI, only to later say that the dip in quality was untenable.

The importance of having a human in the loop is not lost on us, nor is it lost on tech CEOs, who are finally seeing that AI is not ready for the big time just yet.

If recent news has taught us anything, it’s that we no longer need to theorize about the possibility of AI systems going rogue. Recent episodes of AI-enabled attacks have led to warnings from prominent researchers, and calls from frontier labs to slow AI development.

However, AI isn’t just an adversary. It’s also been useful for preventing cyber incidents, which could make it harder to regulate.

All in all, cyber expert Brandon Dixon is unsure about how much slowing down will protect us against AI’s capabilities. He urged businesses to understand their own processes first and map out risks in a recent interview with Tech.co.

Recent AI Attacks Put Safety in the Spotlight

Days before Anthropic CEO Dario Amodei called for frontier labs to slow down AI model development, ex-Anthropic and ex-OpenAI researcher Jacob Coxon released a series of posts on X expressing fear for the future of humanity.

Among the passages, Coxon claimed AI leaders are “gambling with our lives” in their pursuit of superintelligence. His comments most likely would have joined the chorus of existing AI “doomers” as they’re called, had they not followed a series of highly concerning autonomous AI attacks.

Most recently, OpenAI disclosed several concerning behaviors exhibited during the testing of GPT-5.6 Sol. The model supposedly started to leave instructions for future versions of itself, telling it to conceal mistakes and misaligned behavior from users. If that doesn’t concern you, just remember what happened with Hugging Face.

To Brandon Dixon, co-founder and CTO of Ent, which helps organizations understand and protect against malicious actors and AI agents, we’ve moved beyond the theoretical stage of AI cyber attacks. This ultimately means the playing field has shifted massively, and what we do next is important, but also hard to figure out.

AI’s Positive, and Negative, Impact on Security

When I’ve interviewed cyber experts on AI, what we discuss is a kind of double bind. AI is facing both sides of the battlefield, in that it’s helping businesses detect and react to attacks faster, while mounting those attacks in the first place.

“The models being used to craft phishing lures are also being used to find bugs before they ship, improve code quality, and surface vulnerabilities in production systems before they’re exploited,” Dixon told me.

At the moment, it’s impossible to see a world where we fight AI, without using AI. “Speed and efficiency cut both ways,” Dixon told me. Essentially, the features that make AI so good at attacking us, also make it very good at protecting us.

Nevertheless, AI-powered cyber attacks are a serious concern. Using that speed and efficiency, attacks can happen faster, without giving humans enough time to detect them, and serious amounts of data can be lost.

“The limiting factor for attackers has historically been the ability to understand the data that was stolen,” Dixon noted. “But now, AI supercharges that.”

Policing AI is Becoming Increasingly Difficult

Cyber experts have been calling for a more careful consideration of AI’s security vulnerabilities for some time. However, this point in time does feel significant. Dixon said it’s because we’ve now seen these capabilities in action.

“There are documented cases of AI agents breaking out of sandboxes, taking unintended actions, violating corporate policies, or otherwise behaving adversarially. These systems have materialized and evolved extremely quickly,” he explained.

“We are only a handful of years into this technology, and its capabilities change materially every year. That makes it difficult to know where it is heading or to apply security and governance measures that will not become outdated shortly after they are deployed.” – Brandon Dixon, co-founder and CTO at Ent

Despite the technology moving faster and faster, the US government is yet to pass any federal laws around AI. Individual states like California have implemented some AI laws, but currently, AI isn’t equally regulated in the US.

And, given recent comments made by Trump following the slowdown call from AI leaders, it doesn’t look like any new laws will be passed through Congress soon. Mostly, this is because of rapidly developing Chinese AI models, which Trump has continually said the US needs to outpace.

However, many experts will say some government intervention on AI is needed, Dixon included. “I think the government should establish sensible rules that preserve innovation and progress while preventing clearly reckless or harmful uses.”

“The goal should not be to eliminate risk entirely because that would suppress legitimate research and defensive deployment. Oversight should focus on meaningful risks and measurable harms, not broad restrictions on the underlying technology,” he said.

Will Slowing Down AI Protect Organizations?

It’s difficult to think about what happens if we slow down. What’s most important to remember, for starters, is that even though AI leaders have committed themselves to slowing down, they could turn the other way, especially given the stance of the Trump administration.

For Dixon, slowing down AI won’t necessarily put us in a better place. “From my perspective, more time does not necessarily produce a better understanding of security vulnerabilities,” he said, though he does admit he can’t speak for frontier labs themselves.

While additional time could help review code, test software, or identify weaknesses, he added, organizations may not use the time as wisely as we assume they will. “If history is any indication, many will wait for attacks to begin before responding appropriately.”

It’s still not entirely certain what slower development will look like, should it happen. But, Dixon told me businesses shouldn’t use an inability to control all aspects of AI as a reason not to adopt it. To him, it doesn’t make any sense. But what also doesn’t make sense, is just “adopting AI, hoping for the best and not having any clue what people are doing.”

Instead, organizations need to understand how they actually operate. “It is not enough to identify sensitive data, intellectual property, or trade secrets,” he said.

He advises businesses to figure out “which behaviors are acceptable, which workflows deserve attention, how those workflows could be exploited, and how that exploitation would be detected.”

95% of respondents of a new study list AI fluency as a hiring requirement. However, a significant number of organizations are still reporting having made a bad AI hire.

Historically, employees have been encouraged to experiment with AI to find out how it could suit their specific needs. Now, standardized AI use is emerging, which will help hiring managers choose the right talent.

Wouter Durville, CEO of TestGorilla, spoke to Tech.co about what AI fluency means and how businesses can hire with this skillset in mind, all while avoiding common mistakes at the interview stage.

Companies Are Looking for AI-Fluent Job Candidates

TestGorilla’s State of AI Fluency study finds that 53% of hiring managers now prefer a candidate with high AI fluency over one with deep domain expertise. However, of those managers, 59% report having made a bad AI hire.

Wouter Durville, CEO of TestGorilla, highlighted how impactful an AI fluent employee can be. “I think people who are really AI fluent, the outcomes they generate are massive. It’s a big difference.”

Companies certainly want to hire with AI in mind, the study finds, but are often going about it in the wrong way. Durville explains how the study was borne from customers communicating to him how much they were struggling with the issue. Eventually, TestGorilla landed on the term “AI fluency,” one Durville knows is difficult to properly define.

“It’s really about getting those people who know how to use these tools in the right way, and in a responsible way. Who are agile, who are using different tools for different things, who are system thinkers that see the bigger picture of where to use AI and where to put the human in the loop,” he explained.

It goes without saying, though, that this definition will vary based on your company and its goals.

Businesses Are Moving Towards Standardized AI Use

The first step when AI started to make waves in companies was convincing employees to actually use it. Common advice was to let employees figure out the tool for themselves, find ways to embed it into their own workflows, and experiment.

Arguably, this could explain the rise in individual productivity compared to company-wide gains some studies reported. But, this tactic seems to have faded.

“For example, in product development, it becomes quite messy if everyone is using a different tool for coding.” – Wouter Durville, CEO of TestGorilla

Messiness aside, it makes it easier to police the technology if everyone is using it across repeatable, company-wise processes.

That’s not to say organizations should stamp out individual moments of AI brilliance, but standard practice could protect against costly data breaches or common AI mistakes.

In the case of hiring, it’s likely much more productive to have a standardized company practice when it comes to AI. That way, in interviews, hiring managers can assess exactly how a candidate has used AI, and whether it matches up to AI expectations within their organization.

How Can Companies Hire for AI Fluency?

As well as standardizing AI use, there are other tactics companies can use to test AI knowledge. The first is actually defining what AI fluency means to your organization, which 71% of respondents of the survey have already done.

Likewise, the study found 37% of organizations set the bar at tool awareness when hiring a candidate. Rather than putting a tick next to someone’s name when they mention ChatGPT or Gemini, Durville encouraged businesses to “go deeper,” and figure out exactly how the candidate has used AI.

“You want to actually test an example in an interview,” he explained. So, asking a candidate questions like, “tell me about a workflow you actually built, or something that you automated. How did you go about it, what went wrong, and how did you fix it?”

“If you ask those questions, if you peel the onion a bit or go a bit deeper, you’ll very quickly notice, okay, this person has no clue how to do this,” he said.

Furthermore, Durville said AI use — and how it’s going to be tested in interviews — should be related to the job role as much as possible. You’re likely to test a marketing executive on their knowledge of AI video editors, for example, but this is less useful when interviewing for an engineering job.

Overall, Durville has positive views about AI and how it will impact hiring. While the CV and cover letter have basically been killed, he told me, companies can use AI to give more opportunities to candidates at the interview stage.

Recently, AI leaders have been talking about slowing down AI development. In particular, Dario Amodei of Anthropic and has pledged to “pace the frontier,” in a new blog post.

While AI figures such as Amodei, Sam Altman, and Elon Musk appear to agree with each other on pacing, other major players such as Nvidia’s Jensen Huang appear to disagree. Likewise, concerns about Chinese AI one day overpowering US models remain paramount to the US government.

Experts suggest AI leaders should turn their attention to increasing trust, instead of slowing down AI. They also warn that the US slowing down won’t necessarily mean others will do the same.

Anthropic CEO Says ‘Pace the Frontier’

A few days ago, Anthropic’s CEO Dario Amodei released a blog post titled, “We Must Pace the Frontier.” The post is a call for AI leaders to slow down the rapid development of AI models. 

Safety concerns around AI have once again been pulled into headlines following comments made by researcher Jacob Coxon, who resigned from Anthropic amid fears of unrestrained development leading to the end of humanity.

In a post on X, Coxon, who previously worked for OpenAI, claimed the people building AI “earnestly believe it could kill us all by the end of the decade.” Similarly, he claimed leaders are “racing straight to self-improving superintelligence and gambling with our lives.”

AI executives including OpenAI’s Sam Altman and SpaceX’s Elon Musk shared their agreement of Amodei’s post. While the post doesn’t mention Coxon’s recent comments, it does reference the OpenAI/Hugging Face breach.

“The calls from some of the world’s leading AI researchers and technology companies to slow the pace of frontier model development reflect a growing recognition that innovation and responsibility must advance together,” said CISO of Optiv Field, Kristin Lowery.

Lowery suggests accelerating capabilities must come with a significant understanding of how to govern and safely deploy AI, suggesting the call to slow down from leaders could be a positive one.

How Realistic Is Slowing Down AI Development?

It’s one thing for leaders to say they’re going to slow down, but the real question is what exactly this will look like. Amodei’s blog post does outline some strategies for how AI companies can achieve reduced development, and said Anthropic is “unilaterally committing” to one of them. Altman has chimed in to say OpenAI will do the same.

Despite support from Altman, Musk, and Microsoft, which released a new AI code of conduct yesterday, not all leaders agree with Amodei’s plan. Jensen Huang, CEO of Nvidia, for one, made quite a public show of where he stands at the All-In Summit in Los Angeles.

Huang received and answered a call from President Trump on-stage, where the topic was Amodei’s recent post.

“They’re just playing right in the hands of a lot of people that don’t want to see it happen,” Trump went on to say to the crowd. “That could be political people. It could also be China. And we’re not going to let that happen. It’s a hoax.”

“You’re right. We’re not going to let that happen, sir,” Huang responded, and the crowd applauded them.

Chinese AI competition is a major concern for the US government, and this has only ramped up as Chinese models continue to close the performance gap to US models, all while offering a lower prices. The insistence has always been that the US needs to be first ones to crack AI and superintelligence.

From an international perspective, Lowery makes another point about why slowing down could be dangerous. “AI innovation is occurring globally, and not every nation, organization, or threat actor shares the same values regarding transparency, safety, and responsible use. This creates a complex dynamic that increasingly resembles a technology arms race.”

“If responsible organizations dramatically slow innovation while less accountable actors continue advancing without guardrails, we risk creating unintended strategic disadvantages.” – Kristin Lowery, CISO at Optiv Field

Essentially, the US slowing down its own AI development won’t lead to everyone doing the same. Threat actors and hacker organizations could still develop rapidly and cause AI cyber attacks we’re unprepared for.

Why Leaders Should Focus on Building Trust

It’s been hard to trust AI leaders. Whether it’s the constant uncertainty around job losses, the fear of global cyber attacks, or just the end of humanity in general, the AI rhetoric has been far from pretty.

Beyond negative AI messaging, publications are consistently reporting on autonomous hacks and AI data breaches, which suggests we might not be properly prepared to understand and prevent against the dangers AI poses.

Pacing frontier models could be a way to ensure we’re ready to deal with the capabilities of these models. However, experts are saying there’s no reason why we can’t have both safety and high-performing AI.

In a recent interview with Bloomberg, chief ethics scientist at Hugging Face Margaret Mitchell argued safety, privacy, and security must be built directly into AI systems, rather than being seen as a trade off to progress.

Lowery makes a similar point, and says the current debate is the wrong one. “In my view, the debate should move beyond whether to pause or accelerate AI, and focus instead on how to build trust into AI from the outset,” she said.

“History shows that transformative technologies rarely succeed through either unchecked innovation or excessive regulation alone; the most sustainable path is to encourage innovation that drives economic and societal value while implementing thoughtful safeguards that reduce risk and build trust.” – Kristin Lowery, CISO at Optiv Field

Ultimately, AI brings with it more risk the less leaders are willing to pay attention to safety and privacy. Future models should be developed through this lens, so we can reap the full benefits of AI, without risking something far greater.

This article is part of an interview with Betsy Cooper, Director of the Aspen Policy Academy. This interview was conducted for Tech.co’s AI newsletter, The AI Strat. For more interviews with AI experts on the latest trending topics, subscribe to the newsletter.

While the response to AI’s ever-evolving capabilities has been mostly excitement, the threat to cybersecurity in the form of increasing cyber incidents has been difficult to ignore.

Increasingly, experts believe businesses will need to alter internal policies, including incident response plans, to account for the fact that they won’t always be dealing with a human attacker.

I spoke to Betsy Cooper, Director of the Aspen Policy Academy, about how AI continues to change the cybersecurity landscape. Likewise, we discussed how businesses should be prepared to change their practices moving forward.

The AI Threat Hasn’t Been Fully Realized

Before becoming Director of the Aspen Policy Academy, which allows experts in cybersecurity and AI to share their knowledge, Betsy Cooper ran a cyber security program called the Center for Long-Term Cyber Security at UC Berkley. Cooper also worked for the Department of Homeland Security.

When I asked Cooper what excited her about AI, she spoke to its ability to make “it easier to do things that take up a lot of time.” From her perspective, this includes tasks like writing personalized notes to partners.

However, Cooper also has her fair share of concerns, and a lot of them she’s not alone with. She fears what our education system will look like with AI, and how jobs, particularly at the entry level, will change. And of course, she mentions the existential threat of AI.

“Obviously, there’s been a lot of buzz lately about several of the models escaping their confines and hacking others. We’ve talked about that. I wrote scenarios 10 years ago that focused on that idea, and now we’re here,” she told me.

Earlier this week, an Anthropic researcher resigned amid concerns about how the company was handling AI’s threat to humanity. In a series of social media posts, Jacob Coxon wrote: “The people building AI earnestly believe that it could kill us all by the end of the decade.”

While this incident occurred well after my interview with Cooper, it speaks to the concern many experts are feeling. In particular, Cooper voiced concerns for the future of the internet, suggesting we’ll need to ensure it remains valuable to humans in the face of these threats. “I don’t think we’ve fully grappled with that,” she said.

How Incident Response Plans Will Need to Change

Understandably, cyber incident response plans up until now have assumed a human attacker. And, most businesses today will know and have embedded best practice to protect themselves.

However, the introduction of AI agents, which are able to attack autonomously, has changed the game. As a result, Cooper believes internal policies including incident response plans, will need to change with it. In some cases, she says, the humans might be the non-malicious actors, and the agents are the malicious ones.

“That probably requires a different chapter in your incident response plan than the day-to-day. So that would be something that I would really encourage people to do, is start updating your incident response plan to try to figure out the source of the hack, and to figure out if it’s a non-malicious actor human, are you going to use the same playbook that you would as if it were a malicious actor human?” – Betsy Cooper, Director of the Aspen Policy Academy

That being said, Cooper doesn’t see AI as solely a cyber adversary. In many cases, it’s an ally, one that is able to help humans detect unusual activity more effectively.

“Most organizations are going to have to lean into AI to help them identify when something weird is going on, because the patterns may not be as obvious,” she said.

Assess Your Risk Profile and Ensure Basic Security Hygiene

Additionally, Cooper explains how organizations need to be aware of their risk profile when it comes to AI-driven attacks. Hugging Face, the company the rogue OpenAI agents breached, is at a higher risk because of the work it does, Cooper said. Whereas if you’re operating in the paint industry, for instance, your risk is probably lower.

Cooper also expressed the importance of maintaining proper cyber hygiene. AI agents will make the most of shortcuts, if the Hugging Face incident is anything to go by. So, minimizing those and ensuring you aren’t easy to hack is key.

“If you’re not on the easiest path,” Cooper explained, “then you still have hope right now. I can’t promise that’s going to stay that way.”

Ensuring you’re not on the easiest path means sticking to best cybersecurity practices. Cooper gives the example of not using passwords that have already been disclosed, and making sure data is stored in different places and compartmentalized so it can be recovered if you do suffer an attack.

“The same hygiene techniques that hopefully you’ve been already using to prevent the ordinary hacker who’s a human. Many of those techniques at least will make it less likely that you will be targeted,” she said.

However, she does sound a warning about the future of attacks. “Increasingly, this is going to happen more commonly, and it’s probably not possible to stop,” she added. For now, businesses should focus on protecting themselves under the assumption an attack will happen.

Physical AI solutions are, understandably, moving at a staggered pace when you compare them to the breakneck speeds of AI development in general.

A part of this, experts argue, is a trust problem, especially since new physical innovations can lead to more dangerous consequences if underdeveloped or misunderstood. However, there is also the issue of data management and information silos.

I spoke to Sri Teja Kolluri, product lead at Motive, about how fleet owners can upgrade their maintenance processes and ensure success when using AI.

A Trust Gap Remains for Physical AI

When I’ve spoken to logistics experts about the trajectory of physical AI, none of them have been surprised that it’s development has been significantly slower. It’s easy to sit at a laptop and understand the capabilities of Claude, but integrating AI into driver safety, for example, has real world consequences.

Sri Teja Kolluri, product lead at Motive, sees this as a problem that still needs to be solved. “There’s a general excitement about things, but beyond drafting emails, how can we actually use AI in real life?”

Motive’s latest fleet maintenance report suggests fleet owners are trusting AI more than previously. 80% of those surveyed said they would be comfortable with AI summarizing or creating reports, 67% would allow it to schedule or initiate work, and 60% would trust it to approve routine or low-risk tasks.

However, less than half (47%) said they would be comfortable with fully automated workflows without human review, suggesting human oversight remains a critical part of AI workflows.

How AI Can Help Transform Maintenance

Maintenance is one area where Motive and Kolluri feel AI can make a huge difference. Significantly, 80% of survey respondents cited rising maintenance and repair costs as their top operational challenge of the second half of 2026.

Despite the opportunity for AI in maintenance, 60% of owners surveyed aren’t using AI-based predictive or condition-based maintenance tools. Comparatively, 88% of owners are using AI-powered driver safety technology, like AI dash cams and safety programs.

I asked Kolluri about why maintenance may be lagging behind compared to other fleet processes. “With a lot of maintenance workflows, what’s happening is that they’re using different systems that don’t talk to each other. So, you have data scattered across your stack or they’re using spreadsheets, which again, it’s very hard to actually use as input for AI.”

However, Kolluri says maintenance teams are already seeing benefits by opting for a more integrated system. Watco, a transportation and logistics company, reports that problems now appear sooner, more assets stay in service, and it can identify the real cost of operating its fleet.

Hesitant maintenance teams, he continues, should prioritize the “key use cases” that will break down any informational silos in your operation. “I think that’s where you can identify the right solutions to adapt and solve that problem,” he says.

Small Steps and Integration are Key to AI’s Success

Using AI in isolated, random incidents is rarely as productive as using it as part of a connected system. Accountability and ownership are lost, and privacy and compliance decisions are made in reactive bubbles.

Integration and tackling information silos, Kolluri says, are incredibly important to get right when AI becomes involved.

“I think it’s very important that you work towards an AI operation that is well integrated and solves a lot of the data silos.” – Sri Teja Kolluri, Product Lead at Motive

Motive’s report findings confirm the problem. 80% of respondents said their fleet technology environment has some systems integrated, but with significant gaps. Overall, only 13% described their systems as well-integrated that share data automatically across platforms.

As well as integrated systems, Kolluri urges businesses to take small steps forward when introducing AI. “In general, AI might feel scary because, hey, there’s this powerful tool that we can use to automate a lot of things.”

“I think the best opportunity is just start small. Identify these opportunities and then slowly you’ll find that you have built reasonable AI operations,” he says.

Furthermore, don’t start from nothing. AI should be used to accelerate and improve your existing processes. A big part of this is understanding where AI can realistically fit, rather than forcing it.

“Audit your processes, identify where you are spending the most manual effort and where the most repetitive tasks are, and identify those opportunities to see where you can deploy AI. I think that’s the one thing, once you do that you can start taking baby steps towards adopting AI in that specific workflow,” Kolluri says.