70% of small business owners admit to lacking the skills necessary to use AI effectively, according to a new study.

Businesses are, however, becoming more comfortable using AI, and are using the technology more than they did a year ago. 60% of respondents said they use AI every day.

Leaders specifically should focus on finding the correct AI training. Even if they are not using the technology as much as the departments below them, it will help them to respond and relate to teams more effectively.

Small Business Owners Need More AI Training

Thryv’s 2026 AI & Small Business Adoption 2026 report found 70% of small business owners in the US admit to needing more AI training, in order to use the technology more effectively.

“It’s clear that many SMBs are adopting AI faster than they can master it,” Grant Freeman, President of Thryv, stated. “Closing the gap between AI adoption and capability will be critical for Main Street success.”

The survey consisted of 561 small and medium-sized business owners and decision-makers, with most reporting annual revenue of between $1 million and $1.9 million.

AI Adoption Remains High

Despite some uncertainty on how to fully leverage AI’s power, businesses are still adopting the technology at a rapid rate. The study found 66% of US small businesses now use AI, compared to 55% the year before.

Likewise, one third of those surveyed are spending more on AI tools than they were a year ago, and many have already reported gains in both revenue and efficiency. Overall, 70% had increased their revenue, and 55% reported it had helped to reduce costs.

While 46% of SMBs said they would choose AI if it could perform a task equally as well as a human, the study finds AI has not yet directly impacted hiring. Only 13% of respondents said they had hired fewer people because of AI, and 45% expect AI to have no impact on hiring in the next few months.

Proper AI Training is Still Needed for Leaders

It seems AI training is being taken up by different sources. In the study, 31% of respondents turned to other AI tools like ChatGPT to tell them how to use AI. 57% said their primary source of training is YouTube and social media, while 49% cited online resources and webinars.

As adoption still remains high, the study suggests a gap between the rate at which the technology is being introduced, and how quickly it’s being mastered by employees.

“Most small businesses are comfortable using AI. But being comfortable isn’t the same as being effective and that gap is starting to show.” — Grant Freeman, President of Thryv

The main takeaway here is that business leaders should spend time familiarizing both themselves and their departments with AI, before adopting it on a mass scale. This will help prevent against security errors and ensure the technology is making a difference.

Over three quarters of respondents reported individual value findings from AI, compared to just over half of those who reported company-level value, in a new study.

Likewise, the study found that many of the respondents believing themselves to be critical components of company innovation were also creating resistance on AI initiatives.

Viacheslav Brui, Head of AI Transformation at Sombra, spoke with Tech.co about the survey’s findings, and gave his advice for businesses implementing and developing their AI strategy.

Companies are Seeing More Personal Value from AI Than Company-Level

Personal value from AI was more visible than company-level value in a new survey from Sombra. 87% of respondents claimed they had gained transformational, significant, or moderate value from AI in the last 12 months. Comparatively, only 55% of respondents said their company had benefited from AI implementation.

Viacheslav Brui, Head of AI Transformation at Sombra, told Tech.co in an exclusive interview that there are “multiple factors” that can influence whether a company is seeing benefits from AI.

24% of respondents admitted they hadn’t looked into whether their company has benefited from AI implementation, and 15% said leadership had never communicated any gains. As a result, the study suggests local benefits are more visibly seen, whereas company-wide benefits are not always clearly communicated.

The Danger of AI “Champions” and Individual Resistance

The study states, quite plainly, that “innovation has an ego problem.” This is evident in the findings: 79% of respondents saw themselves as a major force behind innovation at their company, and 64% believed innovation would weaken in their absence.

It’s quite common, Brui explains, for companies to select a few AI “champions”, or those that have adopted AI at a very high level. The problem is that this expertise is often not shared with the rest of the organization. “We have one champion, who is on the cutting edge: he knows everything about how to improve performance. But the department in the next room have no idea how to apply that,” Brui says.

The danger of this is that while some employees are seeing individual benefits, the organization as a whole is lagging behind. The best solution is to ensure knowledge is shared across departments.

At Sombra, Brui says, they pick enthusiastic people from each department, give them the tools and the training, and then ask them to share this knowledge with their colleagues. This is better than say, leadership giving general use cases across the business, because it means individuals can adopt AI while remaining “close to their workflows,” meaning overall benefits are far more likely, says Brui.

However, while respondents rated themselves highly on an individual level when it came to innovation, many of the same respondents also admitted to slowing innovation down. 22% had done so because of unreliable data quality, and another 22% said they didn’t see enough value in the project.

Sombra suggests this means innovation depends as much on “individual conviction, control, and willingness to let other people’s initiatives succeed” as it does on organizational support.

How Businesses Can Develop Their AI Strategy

In our interview, Brui broke down what he believed businesses should prioritize when implementing and developing their own AI strategy. He believes leadership should provide support to middle management, and keep them informed on business goals, ownership, and budget.

“Understand what are the company goals and which of those company goals can be achieved with AI, and actually give authority to middle management to decide what initiatives can support these goals.” Viacheslav Brui, Head of AI Transformation at Sombra

Likewise, it’s important for businesses to understand why they are adopting AI. This could be for better ROI or improved customer satisfaction, but “adoption for adoption won’t work,” Brui says.

Another big concern for respondents was data privacy and security, with 58% reporting they’d resisted AI initiatives because of this reason. On this note, Brui says, “I would put in place the policy and the rules of how AI can be used, and set up governance around AI to make sure we prevent data leakage.”

Businesses should also opt for a clear AI policy that details what employees “shouldn’t do with AI so as not to harm the business,” Brui adds.

Overall, businesses should start small. Give employees access to AI tools, and let them understand the basics, like reading documents. Then, maybe introduce AI agents. The main takeaway from Brui? “Don’t build a rocket ship if you cannot build a paper plane.”

This article is part of an interview with Joe Roushar, CEO and founder at Empathi AI, conducted for Tech.co’s newsletter, The AI Strat. For more interviews with AI experts on the latest trending topics, subscribe to the newsletter here. 

AI will eventually lead to the flattening of corporate leadership structures, according to Empathi AI CEO and founder Joe Roushar in an exclusive interview with Tech.co.

Experts weighing in on the subject believe the source of workplace power will change, as AI changes the kinds of capabilities workers have access to. This creates new opportunities to teams on the ground floor and entry-level workers who are new to the workplace.

Despite these changes, Roushar believes in the importance of managers during times of AI investment, and urges businesses to continue to train and monitor management as the technology becomes more and more integrated.

Corporate Structures Will Flatten Because of AI, Expert Says

Joe Roushar, CEO and founder at Empathi AI, believes AI will help “break down the silos” that make hierarchial structures within businesses necessary, leading to a flatter corporate pyramid.

“When you break down the silos, you also break down the stone pipe. Having a flatter structure between the people on the floor and the people in the office will be an inevitable result of more artificial intelligence,” Roushar said in an interview with Tech.co.

With AI, it could be the case that employees below the management level gain access to capabilities that would otherwise have been obtained through years of experience.

How Companies Will Organize Themselves After AI

Anish Batlaw, writing for General Atlantic, describes how he believes workplace hierarchy will more closely resemble a “shape-shifting organism” as AI becomes more widely integrated.

Batlaw explains how tasks will be completed within this new structure: when a task needs to be done, teams come together, organize and use a team of AI agents to achieve the outcome, and collapse when the task is done. “The teams coordinate laterally rather than reporting vertically. They set outcomes, define constraints, and make the judgment calls that require human context.”

Many experts have weighed in on the potential of AI, but it can be intimidating to imagine entire corporate structures changing. Is it even possible to imagine a workplace without managers, and how will ambitious employees develop and rise in their companies without opportunities for promotion?

In an article for Forbes, Bernard Marr suggests AI could provide an opportunity for junior and mid-level employees to come into the workplace with superior skills than those in senior positions. AI will, according to Marr, allow these workers to “produce work that once required researchers, analysts, designers, administrators, and technical specialists.”

This new set up suggests any employee, at any level, will be able to make a large impact within companies. However, it may mean AI will have to be more carefully regulated, if businesses reach a point where senior leaders aren’t interacting with every project AI touches.

Management Remains a Key Influencer

We’re not at the point where managers are obsolete, however, and Joe Roushar explained to Tech.co how he believes management is a vital component of how AI develops and performs within a business.

“They say that the most common reason for people leaving a company is because they have a poor relationship with their supervisor, and I believe that’s true,” Roushar said, suggesting strong management is key to retaining top talent.

“Good companies improve the quality of their middle management. Having AI doesn’t impair that capability, it just makes you have to be more planful and intentional in training your management, in monitoring your management, and in making sure your management continues to incentivize each of your employees who are using AI,” Roushar continued.

Similarly, Roushar advises managers and leaders to continue to empower their employees during times of AI investment, and give them the opportunity to reinvent themselves on their terms. “As AI takes over more brain tasks, and might require fewer people to do those tasks, people will reinvent themselves, and that’s a good thing.”

At the center of Roushar’s philosophy on AI is giving employees the option to be the hero of their own story, and decide what it is AI is going to do for them, not just their company.

Just when you think you’ve learned everything there is to know about AI, a new tool pops up, and it’s back to basics for everyone.

Whatever your role, remaining agile and attentive during this time of rapid AI development is the key to success, and online courses are a great (and free) way to assure you and your business remains competitive.

Businesses are already reporting heavy AI wins. A recent Gallup poll found 65% of US workers say AI has had a positive impact on their productivity. However, another study has found AI performs ‘potentially harmful actions’ 80% of the time, meaning businesses should still exercise caution and remain informed when using the technology.

Building AI Agents with Agno – Edureka

Summary:

• Learn how agentic AI systems work and build your own working agents using Agno’s API
• 4 modules, taking approximately 6 hours to complete
• Beginner-level
• Shareable certificate available on completion

Everybody’s talking about AI agents. Gartner predicts businesses will have over 150,000 agents by 2028, signaling a real demand for elaborate and autonomous systems that live within everyday workflows. So there’s definitely no harm in taking a course that allows you to build your own, especially if you’re yet to do so.

This course from Edureka says you’ll not only be able to build your own working agents, but you’ll learn how to write effective prompts, different tool definitions, and how to structure outputs that make agents more accurate, secure, and automate reliably. Guidance on agentic workflows, model deployment, prompt engineering, AI orchestration, and AI workflows are also included.

The course is recommended for software developers and others with basic experience writing code in Python. Familiarity with APIs, command-line usage, and LLM concepts is also helpful, but most of all the course asks for a willingness to practice through hands-on coding tasks, making it overall great for beginners.

Learn more about this course at Coursera

AI Fluency: Framework & Foundations – Anthropic

Summary:

• Learn how to collaborate effectively with AI
• 14 lectures, 1.1 hours of video, 1 quiz
• Beginner-friendly
• Shareable certificate available upon completion

Anthropic is the well-known makers of Claude, arguably some of the most powerful AI models at the moment. Anthropic offers many different free AI courses, and this one invites users to learn to collaborate with AI “effectively, efficiently, ethically, and safely.”

In this course, you’ll learn The AI Fluency Framework created by Anthropic and professors Rick Dakan from Ringling College of Art and Design and Joseph Feller from University College Cork. This consists of four interconnected competencies: Delegation, Description, Discernment, and Diligence.

Whether you’re deploying AI to your team for the first time, or creating a strong and safe AI culture, this framework promises to help guide human-AI collaboration.

Learn more about this course at Anthropic

AI Prompting for Everyone – DeepLearning.AI

Summary:

• Learn how to effectively prompt AI tools
• 21 video lessons, spanning 7.4 hours overall
• Beginner-level
• Graded assignments and a certificate available with Pro subscription

Companies are spending more and more on AI, and as many AI companies switch to usage-based pricing models, learning how to effectively prompt could make the difference in keeping costs down and sparing your bottom line.

This course, taught by AI pioneer Andrew Ng, provides an up-to-date guide on prompting AI tools for tasks such as finding information, brainstorming and writing, and no-code website building. Aimed at all users, this course provides the fundamentals for any business owner or leader that wants to better their company’s interactions with AI systems.

DeepLearning.ai says the course is a great next step if you’ve already completed its Generative AI for Everyone or ChatGPT Prompt Engineering for Developers courses, but these aren’t essential to starting.

Learn more about this course at DeepLearning.AI

Ethical Considerations for Using Generative AI – IBM SkillsBuild

Summary:

• Learn the fundamental ethical considerations of using AI
• 60-90 minutes total course time, available in 7 languages
• Beginner-level
• Users must be registered to take part

IBM is a champion of AI courses, and has plenty for registered members to complete for free.

On this course, you’ll “explore the fascinating and complex world of ethics as they relate to generative AI.” Likewise, you’ll discover the specific risks related to data inputs and consider the ethical implications of AI-generated output. Included is also guidance from the IBM AI risk atlas.

Business leaders should be fully aware of the ethical implications of any AI investments. By learning these fundamentals, businesses can protect themselves from legal and reputational risk, and ensure they remain compliant with regulations while using the tools.

Learn more about this course at IBM

Why It’s Important to Keep Learning AI

Leaders might no longer be predicting a jobs apocalypse in response to AI, but the technology is still running rampant in workplaces. The companies exposed to AI are seeing as much as a 163% growth in productivity in some studies, and AI spending continues to rise.

By focusing on what they have learned previously with AI and failing to keep up with new developments, business leaders risk being left behind. It’s hard to predict where the technology will be even in the next year, so taking advantage of free courses like the ones above means leaders can correctly guide their teams through the next AI wave.

Even if you’re not using these tools yourself, having an understanding of how they work will help inject a positive, AI-friendly work culture into your company.

China has finally caught up with Mythos on cybersecurity. Zhipu AI, known as Z.ai, has released its open-weight GLM-5.2, and experts are concerned that it could match up to Claude’s Mythos, which is considered the gold standard in cybersecurity.

While it reportedly lags behind in general areas, China appears to have closed the gap between itself and the US in the realm of defense, with Z.ai demonstrating a similar level of proficiency in detecting security vulnerabilities.

The news will cause panic to echo through the halls of power. The US government has gone to great lengths to restrict China’s access to Mythos and related models, with such tools regarded as serious threats to national security.

China Rivals Mythos on Cybersecurity With New AI Model

China’s Z.ai has released its latest AI model, GLM-5.2, which can detect security vulnerabilities as effectively as Mythos, researchers claim. While it can’t compete with the likes of Claude and ChatGPT in terms of general performance, it has proven highly capable in “cybersecurity scenarios.”

Revealed in April to significant alarm, Mythos is a tool for detecting vulnerabilities in cybersecurity infrastructure. US lawmakers have warned that it could instead be used to power a catastrophic wave of cyberattacks.

In the interim, the federal government has been engaged in a protracted saga with Anthropic surrounding public access to the platform. Recently, it granted the company permission to release its latest model, Mythos 5, to a group of about 100 businesses and federal agencies.

Cheaper & Publicly Available

Z.ai GLM-5.2 is an “open-weight” model, which means that it can be downloaded and operated by anyone with the relevant hardware. It can also be modified, with bad actors theoretically able to remove safety guardrails and use it for illicit purposes.

“An attacker can run it locally without safety guardrails, fine-tune it against their specific targets, and operate with zero visibility to any provider or defender.” – Travis Lanham, CTO and founder of Armadin

Compounding these anxieties, GLM-5.2 is also significantly cheaper than comparable US models, including Claude Opus 4.8 and ChatGPT-5.5. There are reports that it could be an “illegal distillation” of those two models, with Anthropic earlier this year expressing concern that Chinese companies were launching attacks to steal its model weights.

Businesses Must Act Now to Avert Security Disaster

There are fears that the repercussions could be catastrophic. Business cybersecurity is already in a dire position. As attackers harness AI to commit increasingly sophisticated data breaches, firms suffer the impacts of a yawning skills gap. If it falls into the wrong hands, GLM-5.2 could make a bad situation infinitely worse.

Businesses need to overhaul their cybersecurity practices. As a starting point, they need to invest heavily into the best talent. With higher education facilities across the US doing their bit to close the skills gap, this should hopefully become easier in the near-future.

At the same time, firms should focus on training their existing workforce on how to spot cyberattacks. This is a problem that starts at the top, with a scarcely believable 98% of bosses unable to identify all the signs of a phishing attack.

And finally, it’s critical that companies adopt the latest technology. There is cause for optimism in this respect, with 83% of businesses already using AI to combat data breaches. As AI models continue to develop apace, so, too, do answers to the questions posed when powerful technology falls into the wrong hands.

AI regulation is apparently a bipartisan issue, with a new study showing that both Democratic and Republican voters are heavily in favor of more robust standards for the technology.

The report pointed to a wide range of measures that AI companies should address, like a guaranteed “off” switch, regulatory guardrails, and pre-release safety tests.

In 2026, businesses need to consider the fact that everyday users are still quite skeptical about AI and want to see it reigned in before it’s too late.

AI Regulations Are a Two-Party Issue

According to a report titled The AI Safety Majority from the Artificial Intelligence Policy Institute (AIPI), Democratic and Republican voters in the US believe that AI regulation is important.

More specifically, 82% of Democrats and 77% of Republicans believe that the industry needs regulatory guardrails to remain safe.

This report surveyed 1,007 U.S. likely voters across the country to collect data in regard to how Americans feel about AI.

What Kind of AI Regulation Do Americans Want?

Democratic and Republican voters weren’t just aligned when it came to applying required guardrails for AI regulation. Across the survey, the party voters were mostly on the same page when it came to various flagship items of AI regulation, including:

• Guaranteed off switch: 88% D vs 83% R
• Congress address loss of control: 87% D vs 82% R
• Don’t build until controllable: 84% D vs 83% R
• Mandatory pre-release safety tests: 72% D vs 72% R

Simply put, there are very few issues on which these types of voters agree, but when it comes to regulating AI, the partisan gap is a lot smaller.

AI Concern Remains High

In 2026, AI is not the silver bullet many businesses hoped it would be. AI errors from simple typos to full-on database deletions have eroded public trust in the technology so much that most people are more concerned than excited.

The data from AIPI found that people are far more concerned about the negative impacts of AI (56%) than they are excited about the changes from its growth (18%).

All that to say, businesses need to be cautiously hesitant when it comes to diving all-in on AI technology. Without a human-led strategy, you could end up with a lot more problems than solutions.

New research from Pegasystems finds the businesses seeing the most success from agentic AI implementation are focused on reimagining existing processes. This suggests businesses should aim to enhance existing functions based on their needs.

Peter van der Putten, director of the AI Lab at Pegasystems, spoke to Tech.co about the most meaningful measure of agentic success being “adoption at scale.” Similarly, he adds, “Organizations must connect AI initiatives to strategic priorities and core business processes and tie them directly to measurable outcomes and metrics.”

The study urges businesses to examine their existing processes, sourcing the ones that can be automated, and use collaborative and innovative approaches to maximize deployment based on clear governance and business goals.

Use Agentic AI to Reimagine Existing Processes

New research from Pegasystems Inc. and research firm Savanta, finds organizations seeing the most success when delivering agentic AI are doing so by rethinking the existing processes they already have in place.

Overall, 96% of respondents said they had achieved success by rethinking existing processes. 53% said they had done this to a “significant” extent, by reimagining everything their company does.

For this report, Pega surveyed 500+ business and IT decision makers worldwide who had successfully implemented agentic AI projects.

Support a Culture of Collaboration and Innovation

The businesses seeing the most success with agentic AI are fostering a culture of collaboration and invention. 80% of respondents report business and IT operations were willing to introduce new technology, innovation, and ideas to explore new possibilities. Most of this stems from a desire to take existing processes and make them more predictable and consistent.

This doesn’t necessarily mean simply replacing entire workflows with agentic capabilities, however. Peter van der Putten, director of the AI Lab at Pegasystems, explained in an interview with Tech.co that human and AI collaboration remains a critical element. 

“Agentic AI should not be pursued in isolation,” van der Putten says. “It must be combined with other forms of artificial and human intelligence and orchestrated within how work is executed across the organization. This includes both straight-through workflows and human-in-the-loop decision-making.”

Similarly, van der Putten expresses the importance of governance and explainability. “Agent behavior must be auditable, measurable, monitored, an controlled to ensure consistent and predictable outcomes.” This not only creates more consistent outcomes, but protects businesses from common AI mistakes like poor data quality.

Prioritize Corporate-Level Strategy

A staggering 95% of businesses who successfully deployed AI agents had a specific, corporate-level strategy and plan for execution. This strategy should be tied directly to success metrics, which are regularly reviewed and evaluated based on progress.

In a press release, chief technology officer at Pegasystems Don Schuerman said, “We’re fast reaching a tipping point with agentic AI where adoption is high within organizations, but maturity is not.”

Curious about what Schuerman believes “mature” AI adoption looks like, I asked van der Putten, who started by defining what it doesn’t look like. He dismissed “large data science teams experimenting in isolation,” and simply providing employee access to tools like Claude or Gemini.

Instead, it “reflects a shift from experimentation to execution, from activity to outcomes, and from isolated agents to agentic AI embedded in orchestrated systems of work.” To achieve this, van der Putten adds, leaders must align business and IT around shared outcomes, and design for “predictability, governance, reduced complexity, and scalability from the outset.”

How Businesses Should Focus Agentic Investments

In terms of what businesses should focus primarily on when deploying agentic AI agents, van der Putten said: “Too often, core and primary business functions, especially those running on legacy systems, are left untouched. This is understandable, as these environments consume the majority of IT resources and governance attention. However, it is also a critical mistake.

“Agentic AI should be used to reimagine these core systems and processes as part of a modern AI-enabled enterprise architecture. The starting point must always be the processes and outcomes that matter most to the business, rather than isolated use cases at the periphery.”

The message for businesses is to be strategic, but also, think holistically. Gaining a strong understanding of how individual processes connect to wider success, sets you up for an agentic solution that supports all business functions.

You can now officially talk to Claude in Slack, with Anthropic announcing the new Claude Tags feature for messaging platforms.

By giving Claude access to their Slack platform, users will be able to tag the AI to ask questions, request specific tasks to be performed, and receive updates on company priorities.

Claude Tags is currently only available for Enterprise and Team users, and it’s only available in beta, but the company obviously has plans for a wider rollout in the future.

Anthropic Unveils Claude Tags

Announced in a company blog post, Claude Tags are a new way to access the increasingly popular AI model on messaging platforms like Slack.

Users will be able to tag @Claude in Slack conversations to get information and perform tasks without having to leave the messaging platform.

Claude Tags can currently be used in Slack by Team and Enterprise users in beta, although the company has plans to roll out the feature to broader Claude users and other messaging platforms.

What Can Claude Tags Do?

Claude Tags can perform a wide range of tasks for your staff, without your employees having to leave the messaging platform at all. It can:

• Work collaboratively – Claude Tags exist in Slack channels for everyone to see, so your team can collaborate on what is said and how to move forward.
• Provide updates – Claude Tags learn over time in specific Slack channels, so it can provide relevant updates.
• Perform tasks – You can instruct Claude Tags to perform specific tasks right in Slack and follow its progress in threads.

Claude Tags will replace the existing Claude for Slack app in the messaging platform.

Is Claude Best for Business?

In the early AI years, ChatGPT was the go-to solution for the majority of businesses, because it was the most advanced and accurate among the various chatbots.

However, over the last few months, Claude has established itself as the clear favorite, with more and more businesses getting on board. This is largely because of Claude Code and Claude Cowork, which makes automating businesses tasks infinitely easier.

With the launch of the decidedly convenient Claude Tags, Anthropic could be positioning itself as the de facto business AI for the foreseeable future.

The AI boom has had a negligible impact on the US workforce and wages, according to a new study from the European Central Bank (ECB). Despite firms’ heavy investment in the technology, fears that AI will lead to mass unemployment are so far largely unfounded.

According to the research, the US economy began to adjust to the emergence of AI several years ago, with jobs from the most vulnerable sectors being reallocated to other segments. This has gradually reshaped the labor market, but the study notes “no major income effects” from this shift.

Researchers do not rule out more pronounced changes in the future, however. As CEOs across the country gear up to make layoffs in favor of automation, businesses should be mindful that a winning AI strategy is built on maintaining a “human-in-the-loop.”

AI Boom Yet to Significantly Impact US Workforce and Wages

The ECB has found that the AI boom of the last few years has not yet substantially altered the US workforce, despite fears to the contrary. There has also been no noticeable impact on wages, the report notes.

Some workers may be displaced, with junior staff in “highly exposed sectors” thought to be most at risk. However, the report observes, the overall aggregate impact on the workforce and wages has so far been minimal.

The US economy begun its adjustment to the AI explosion several years ago. In the interim, jobs from the most vulnerable sectors have been reallocated to other segments, which has slowly reshaped the labor force.

Workforce and Wages Largely Unchanged Since 2019

According to the ECB: “All else being equal, between 2019 and 2025, jobs with a high substitution risk grew by around 15 percentage points less than jobs with a low substitution risk.”

Across this period, jobs with a “high substitution risk,” such as economists or graphic designers, declined by an average of more than 4%, while low-risk jobs, including electricians and teachers, increased by 13%. On a macro level, low-risk jobs grew from 23% to 25%, while high-risk jobs dropped from 35% to 33%.

ECB has observed no major impact on US wage, stating: “AI substitution risk has had no significant impact on wage growth since 2019.” It did, however, sound a note of caution: “Over time, as the labor market continues to adjust and AI tools become more generative, income effects may be more pronounced.”

Successful AI Strategy Hinges on Human Involvement

With the majority of US businesses now leveraging AI, and many more set to embed it into their ways of working in future, firms face renewed pressure to develop a rigid deployment framework. When a company rushes to embed AI without pausing to consider the ramifications, it can lead to “AI debt,” which can have damaging reputational and financial impacts.

In order to mitigate this, it’s crucial that businesses reflect on their AI strategies before they pull the trigger. While the rewards can be massive, the inverse is also true — failure to properly map out your integration plan will only hurt your prospects.

Human involvement is a key part of this. The temptation is to outsource wholesale processes to technology, but keeping a “human-in-the-loop” will ensure that you avoid obvious mistakes, leave the door open for data breaches, and ultimately waste vast amounts of time and money.

Google’s AI lab, DeepMind, has just introduced a new security framework that classifies autonomous, untrusted artificial intelligence agents as internal security concerns, calling them “potential insider threats.”

The roadmap is designed to contain and monitor AI systems that try to bypass human oversight.

At the same time, OpenAI is rolling out a new raft of security features for its own AI tools. In 2026, it seems, top AI companies are taking security risks more seriously than ever.

How Does Google’s New Roadmap Work?

Google Deepmind announced their “AI Control Roadmap” in a blog post that revealed a rough overview of the framework.

It covers “system-level mitigations that limit the harm a misaligned AI system could cause” across four different areas. First is threat modeling, which lists tactics to use against a “hypothetical AI adversary,” assuming a worst-case scenario.

The other areas covered in the roadmap include two control invariants (specifically, reliable detection of misaligned intent and effective responses to attack attempts), capability-based mitigation ladders (defensive measures that can evolve alongside AI model capabilities), and a portfolio of practical defenses, which is a list of specific, actionable interventions or safeguards.

If you want even more information on all of the above, check out Deepmind’s full 36-page research paper.

Deepmind’s Mitigation Plan

The same announcement also discussed Deepmind’s plan to “deploy AI control mitigations” through a two-pronged approach.

• Identifying the problem (Detection): Uses trusted AI systems to constantly review the less-trusted working agent’s reasoning, actions, and plans in order to keep them on track.
• Managing the problem (Prevention and Response): This gives supervisors their next steps, should they spot an agent taking harmful actions.

While all of these responses might appear to be validating fears of a robot uprising straight out of Terminator or The Matrix, it’s actually a more mundane response.

This new roadmap is essentially taking a zero-trust approach to computer security: We can only be sure that we’re preventing unauthorized system actions by never actually trusting AI agents to be fully autonomous, regardless of how likely they are to go off book with harmful actions.

OpenAI Boost Security Protocols As Well

It’s also worth mentioning that OpenAI has itself announced several security-oriented changes this week.

The company is releasing an updated iteration of the security-focused model GPT-5.5-Cyber and has just released the Codex Security scanner as an app plugin, giving users a simple way to vet code in order to identify, validate, and patch any vulnerabilities.

OpenAI’s ChatGPT and Google’s Gemini are two of the most well-known AI tools operating today.

Seeing both companies working towards increasing security measure surrounding AI agents and tools is an indicator that the cybersecurity ecosystem is well aware of the many challenges that it currently faces as LLMs continue expanding their influence to, well, just about everywhere.