AI Is Moving Faster Than Businesses Are Adapting. Here's How to Close the Gap

Ask a room of business leaders whether their company uses AI, and almost every hand goes up. In McKinsey's 2026 global survey, nearly nine in ten respondents said their organization uses AI regularly in at least one part of the business, and 44% said it was now scaling across the enterprise (McKinsey, 2026↗ (opens in new tab)).

Now ask a harder question: is it paying off, and is it under control?

The answers are less comfortable. Only 37% of those respondents could point to any effect on their company's profits, a share that hasn't moved in a year. Among large US public companies surveyed by EY in mid-2026, 98% had formal AI policies, yet 47% admitted skipping them when a deployment felt urgent. A quarter of those using AI agents said they couldn't detect agents nobody had approved (EY, 2026↗ (opens in new tab)).

That is the real story of AI in business in 2026. Adoption is no longer the hard part. Adaptation is: the measurement, governance, security, skills and accountability that turn a powerful tool into a safe and profitable one. And as AI moves from writing answers to taking actions, the price of that gap is rising.

This guide separates evidence from forecast and hype, shows where AI is creating real value and where it is creating real exposure, and ends with a framework and a week-one plan for businesses of any size.

AI has arrived. The returns haven't, yet.

AI is now routine inside most large organizations, but most of them can't yet show it's improving their profits. The distance between those two facts is the most important thing to understand about AI in business in 2026.

Why “88%” and “one in five” can both be true

You'll see two very different answers to "how many businesses use AI?" In McKinsey's 2025 global survey, 88% of organizations used AI in at least one business function, and 70% used generative AI in at least one (Stanford AI Index, 2026↗ (opens in new tab)). Yet the US Census Bureau, which surveys a representative sample of all US employer businesses, found AI in use at only 17% to 20% of firms between December 2025 and May 2026 (US Census Bureau, 2026↗ (opens in new tab)).

Both numbers are accurate. They describe different populations. McKinsey's respondents lean toward larger organizations, and the question is whether AI is used anywhere in the business. The Census sample is dominated by small firms, and it asks whether AI is used in actually producing goods or running business functions. Size matters: 37% of US firms with 250 or more employees used AI, compared with fewer than one in five of those with under 20.

For a small business owner, that's a useful corrective. If it feels as though everyone else has AI figured out, they mostly don't. Plenty of owners use a chatbot themselves. Far fewer have built AI into how the business actually runs.

Adoption is rising. Measurable value isn't keeping up

Inside the organizations that have adopted AI, use is deepening fast. In McKinsey's 2026 survey, the share saying AI is scaling across the enterprise rose from 38% to 44% in a year, and 56% now use it in three or more functions. Eight in ten respondents say AI has improved their own productivity (McKinsey, 2026↗ (opens in new tab)).

The money is moving even faster. Corporate investment into AI companies more than doubled in 2025, to $581.7 billion. Note what that figure is: deal value flowing into AI firms, not what other businesses spend on using AI.

Profit is the part that isn't moving. Just 37% of McKinsey's respondents attribute any positive effect on earnings to AI, essentially the same as a year earlier, and only about 6% are "high performers" that credit AI with 5% or more of their earnings. McKinsey's own summary is that organizations' conviction in AI is growing faster than the returns they can attribute to it.

We call that the AI adaptation gap: the distance between how widely an organization uses AI and how far it has rebuilt the measurement, governance, security, processes, skills and accountability needed to use it safely and profitably.

“Adoption is no longer the hard part. Adaptation is.”

None of this means AI is overhyped everywhere. Where the work is well defined, the evidence of value is some of the strongest in modern management research.

Where AI is creating real value

AI's clearest, best-measured gains come from bounded tasks where a person can quickly check the result: answering support tickets, writing code, drafting documents. The gains are real, often large, and uneven.

The strongest evidence: bounded, checkable work

Some of the most reliable evidence on AI at work comes from field experiments, where some employees get an AI tool and comparable colleagues don't. When a Fortune 500 software company rolled out an AI assistant to 5,179 customer-support agents, issues resolved per hour rose about 14% on average (Brynjolfsson, Li & Raymond, 2025↗ (opens in new tab)). Across randomized trials covering 4,867 developers at Microsoft, Accenture and a Fortune 100 company, an AI coding assistant raised completed tasks by about 26% (Cui et al., 2025↗ (opens in new tab)).

Writing shows a similar pattern. In a controlled experiment with 453 professionals, ChatGPT cut the time spent on writing tasks by 40% and raised quality ratings by 18%, though the tasks didn't require factual precision or company knowledge (Noy & Zhang, Science, 2023↗ (opens in new tab)).

Why less-experienced people and small teams can gain more

The most consistent pattern across these studies is who benefits. In the support study, the newest and least-skilled agents improved by about 34%, while the most experienced barely changed. The coding trials found the same skew toward less experienced developers. The researchers' explanation is that AI captures and redistributes the tacit know-how of a firm's best people, helping newcomers move up the learning curve faster.

The jagged frontier: where help turns into harm

The gains have an edge, and it isn't visible from the inside. In a preregistered experiment with 758 Boston Consulting Group consultants, those using AI on tasks within its capabilities completed 12.2% more work, 25.1% faster, at over 40% higher quality. On a task that looked similar but sat outside what the AI could do well, consultants using it were 19 percentage points less likely to get the right answer than those working without it (Dell'Acqua et al., 2023↗ (opens in new tab)). The researchers called this boundary a "jagged frontier".

People are also poor judges of their own AI-assisted productivity. In a 2025 trial by the research group METR, experienced open-source developers took 19% longer with AI tools while believing they'd been 20% faster (METR, 2025↗ (opens in new tab)). METR has since said developers are probably sped up by newer tools, but that its 2026 data was too affected by who chose to take part to measure by how much (METR, 2026↗ (opens in new tab)).

And task gains don't automatically become economic gains. A study linking AI chatbot use to official Danish payroll records across 25,000 workers found no measurable effect on earnings or recorded hours two years after adoption (Humlum & Vestergaard, 2025↗ (opens in new tab)). That's consistent with McKinsey's finding: individuals feel faster long before companies book the profit.

The two sides aren't mirror images. Value is proven in a handful of well-defined places. Exposure is broader and more ordinary than most headlines suggest.

Where the real risks are

The most common AI risks in 2026 aren't exotic. They are confident errors that reach customers, sensitive data drifting into tools nobody approved, and criminals using the same technology to impersonate people you trust.

Wrong answers that look right

AI systems still produce fluent, confident mistakes. When 22 public broadcasters in 18 countries had journalists assess more than 3,000 answers about news from ChatGPT, Copilot, Gemini and Perplexity, 45% had at least one significant issue: 31% had serious sourcing problems and 20% major accuracy problems (EBU/BBC, 2025↗ (opens in new tab)). Those were news questions, not business tasks, but the failure mode is the same: the wrong answer looks exactly like a right one.

The consequences land on the business, not the model. In February 2024, a Canadian tribunal held Air Canada liable after its website chatbot invented a refund rule for bereavement fares (BC Civil Resolution Tribunal, 2024↗ (opens in new tab)). In October 2025, Deloitte agreed to partially refund the A$440,000 the Australian government had paid for a report that contained a fabricated court quote and references to research that didn't exist (AP via CTV News, 2025↗ (opens in new tab)). The damages in the Air Canada case were small. The principle isn't: a business owns what its AI says.

Data that leaves without anyone noticing

Shadow AI is the use of AI tools, accounts or integrations that an organization hasn't approved or can't see: an employee pasting a client contract into a personal chatbot account, a browser extension that reads email, a meeting recorder nobody signed off. It usually starts with good intentions. People use AI because it helps them, and the approved option is missing or slower.

It is also expensive when it goes wrong. In IBM's 2025 study of 600 organizations that suffered data breaches, one in five had a breach involving shadow AI, which added about $670,000 to the average cost. Of organizations that had a breach involving their AI models or applications, 97% lacked proper AI access controls (IBM, 2025↗ (opens in new tab)). Security vendors that monitor company networks consistently report employees sending sensitive data to AI apps, often from personal accounts the company can't see.

Attackers have AI too

AI-assisted crime is now visible in official numbers. In IBM's 2026 study of 602 breached organizations, one in four malicious breaches was AI-enabled, mostly through deepfake impersonation and AI-enabled malware, and those breaches cost about $6 million on average, roughly $1 million more than the typical breach. Organizations using AI and automation in their own security operations cut breach costs by almost $2 million (IBM, 2026↗ (opens in new tab)).

In the FBI's 2025 internet crime report, which included a section on AI for the first time, the FBI received 22,364 complaints citing AI, with reported losses of nearly $893 million (FBI, 2026↗ (opens in new tab)). Reported losses understate the real total, since many victims never report, or never realize AI was involved.

Attackers are getting faster, not magical. Google's threat intelligence team reported in September 2026 that attackers had used an agent-based framework to run a mass credential-harvesting campaign in under six hours. The same report said it had not yet seen attackers deploy fully autonomous pipelines that discover and exploit new vulnerabilities on their own (Google Threat Intelligence Group, 2026↗ (opens in new tab)). For most businesses, the realistic threat is better phishing, cloned voices and stolen logins.

Copyright, reputation and the rules

The law is unsettled, and it's moving. In the US, purely AI-generated work can't be copyrighted: the Supreme Court declined in March 2026 to hear a challenge to the human-authorship rule (Mayer Brown, 2026↗ (opens in new tab)). On September 29, 2026, a federal appeals court ruled for the first time that training an AI tool on copyrighted material, in this case legal research summaries used to build a competing product, wasn't fair use (Insurance Journal/Reuters, 2026↗ (opens in new tab)). The court expressly left open the bigger question of training generative AI, and the losing company says it will ask the Supreme Court to review the decision (LawNext, 2026↗ (opens in new tab)).

Regulation, meanwhile, slowed down in 2026 rather than speeding up. The EU's Digital Omnibus, in force since July 2026, pushed its strictest "high-risk" AI rules back to December 2027, but the AI Act's transparency rules on chatbots and AI-generated content began applying in August 2026, with some labeling duties following in December (Orrick, 2026↗ (opens in new tab)). The US has no comprehensive federal AI law, and the federal government is challenging state rules (Sidley, 2025↗ (opens in new tab)); Colorado repealed its landmark AI Act in May 2026 and replaced it with a narrower law (Goodwin, 2026↗ (opens in new tab)). The UK still relies on existing law and sector regulators. None of this is legal advice, and the details change often.

The practical takeaway: don't wait for regulators to define safe AI use for you. Disclosure is becoming the closest thing to a global baseline, and your customers will judge your AI use long before any regulator does.

Every risk so far shares one assumption: a person reads the output before anything happens. AI agents change that.

From answers to actions: what AI agents change

An AI agent is an AI system that pursues a goal by deciding its own steps, using tools such as email, files, browsers and software, and acting on the results. That makes agents more useful than chatbots, and their mistakes more costly.

What an AI agent actually is

Anthropic, which makes the Claude models, draws a useful line between "workflows", where AI and tools follow steps a developer has fixed in advance, and "agents", where the AI dynamically directs its own process and tool use (Anthropic, 2024↗ (opens in new tab)). Many products sold as agents are really the first kind. Gartner has a name for the rebranding of chatbots and automation as agents, "agent washing", and estimated in 2025 that only about 130 of the thousands of vendors claiming agentic AI were offering the real thing (Gartner, 2025↗ (opens in new tab)).

What agents can do now

Capability is climbing quickly on benchmarks. Stanford's 2026 AI Index reports agents' success on one real-world task benchmark rising from 20% in 2025 to 77.3%, and on cybersecurity challenges from 15% in 2024 to 93% (Stanford AI Index, 2026↗ (opens in new tab)). Benchmarks aren't workplaces, which have messy permissions, ambiguous goals and people trying to trick the agent, but the direction is clear.

Large companies are moving fastest. Among McKinsey's respondents at firms with more than $1 billion in revenue, the share scaling AI agents in at least one function rose from 27% to 40% in a year, while smaller organizations stayed flat at 22% (McKinsey, 2026↗ (opens in new tab)). In EY's survey of large US public companies, 91% of senior AI leaders said their organization uses agentic AI in pilots or full deployment (EY, 2026↗ (opens in new tab)).

When the mistake is an action

The risk doesn't only come from the agent's own mistakes. Agents read emails, documents and web pages written by other people, and instructions hidden in that content can redirect them, a technique called prompt injection. In 2025, researchers disclosed "EchoLeak", a flaw in Microsoft 365 Copilot where a single crafted email could make the assistant pull internal data into an outbound link without the user clicking anything. Microsoft fixed it and there was no evidence of real-world exploitation (EchoLeak paper, 2025↗ (opens in new tab)), but the lesson stands: an assistant connected to your systems inherits every permission it touches.

The governance gap

Controls are lagging the rollout. In EY's survey of large US public companies, 98% had formal AI governance policies, yet 47% said those processes had been skipped for urgent deployments. Of companies using agentic AI, 49% hadn't updated their governance to cover agents, and 26% couldn't detect unauthorized agents operating internally. More than a third, 36%, had already experienced an AI incident with a materially negative impact.

Agents change who does the work. That brings us to the question most people actually care about.

Will AI replace jobs, or redesign them?

So far, AI is changing tasks and entry-level hiring faster than it is eliminating whole jobs. The evidence for large-scale job losses hasn't arrived. The evidence of change at the bottom of the career ladder has. It helps to keep three kinds of evidence apart: what has happened, what employers expect, and what researchers forecast.

What has happened

Across the US economy, there's no clear sign of AI-driven disruption yet. Yale's Budget Lab, tracking monthly government labor data through August 2026, finds that measures of AI exposure and usage show no clear relationship to changes in employment or unemployment (Yale Budget Lab, 2026↗ (opens in new tab)). In Denmark, where researchers could link chatbot adoption to payroll records, they found no measurable effect on earnings or hours two years on.

Inside companies, the cuts have been smaller than predicted. In McKinsey's 2025 survey, 32% of respondents expected AI to reduce their organization's headcount over the following year. In 2026, 14% of respondents at AI-using organizations said it actually had, and two-thirds reported little or no AI-related change in total employment (McKinsey, 2026↗ (opens in new tab)).

What employers expect

Expectations are rising anyway. In McKinsey's 2026 survey, 39% of respondents expect AI to reduce their organization's headcount in the coming year, while 43% expect little or no change. Only 13% say AI makes them anxious about their own career. Employer surveys have overestimated near-term cuts before, so treat expectations as intentions, not outcomes.

What researchers forecast

The International Labour Organization estimates that one in four workers worldwide is in an occupation with some exposure to generative AI, but only 3.3% are in the most exposed category, and women are more exposed than men. Its conclusion is that transformation of jobs is far more likely than outright replacement (ILO, 2025↗ (opens in new tab)).

The first rung of the ladder

Still, the pattern fits what the productivity studies would lead you to expect. If AI helps novices most, the tasks most easily handed to it are the ones juniors used to learn on. Stanford's AI Index reports that employment of software developers aged 22 to 25 has fallen nearly 20% since 2024, even as older developers' headcount grew (Stanford AI Index, 2026↗ (opens in new tab)).

For employers, that creates a quiet long-term risk: if nobody does the junior work, nobody becomes the senior expert. It's also a reason to measure before cutting. In August 2025, Australia's Commonwealth Bank reversed 45 job cuts it had linked to a new voice bot after its union showed call volumes were rising, and the bank admitted its assessment had been wrong (Bloomberg, 2025↗ (opens in new tab)).

What AI means for brands and discovery

AI makes content cheap to produce and search answers instant, which shifts the advantage toward brands that are distinctive, credible and easy to cite. Search isn't disappearing. How people get from a question to a business is changing.

More content, less difference

A 2024 experiment in Science Advances gave some writers story ideas from an AI model. Their stories were rated more creative and better written, especially among less creative writers. But the AI-assisted stories were also more similar to each other: individually better, collectively narrower (Doshi & Hauser, 2024↗ (opens in new tab)). Audiences can also react to the label itself. In experiments with more than 1,000 US adults, simply mentioning "artificial intelligence" in product descriptions lowered emotional trust and purchase intent, especially for high-risk purchases (Washington State University, 2024↗ (opens in new tab)).

In practice, that means using AI for volume and variation, and keeping humans in charge of the ideas. Start from a clear position in your market and content strategy, not just content production, then test creative instead of guessing which AI-assisted variations actually perform.

Search isn't dying. Clicks are moving.

Google said in May 2026 that its search queries had reached an all-time high, while its AI Overviews reach more than 2.5 billion people a month and AI Mode passed a billion monthly users within a year of launch (Google, 2026↗ (opens in new tab)). People are searching more, not less. What's changing is what happens next. In Pew Research Center's analysis of 900 US adults' browsing, people clicked a result on 8% of Google visits that showed an AI summary, compared with 15% when there was no summary (Pew Research Center, 2025↗ (opens in new tab)).

The next layer is agents that act. At its 2026 developer conference, Google showed Search agents that monitor the web, book services and call businesses on a user's behalf (Google, 2026↗ (opens in new tab)), and in September 2025 ChatGPT began letting US users buy from participating merchants without leaving the chat (TechCrunch, 2025↗ (opens in new tab)). How many purchases flow through these agents isn't public yet.

What brands should do now

There's no switch that guarantees a place in an AI answer, and Google says its normal search fundamentals still apply to AI Overviews. We've covered how AI search visibility actually works in detail elsewhere. The short version:

• Be the source worth citing. Publish specific, verifiable information: prices, policies, original data, real expertise.
• Keep facts consistent and machine-readable everywhere: your site, your Google Business Profile, directories and product feeds. Agents can only book or buy what they can read.
• Don't abandon technical SEO. AI Overviews are built on Google's normal crawling and indexing.
• Build demand for your name. When AI handles generic questions, people searching for you specifically matter more.
• Measure leads and revenue, not raw sessions. Fewer, better-qualified clicks can still mean more business.
• Set AI content standards: what AI may draft, what a person must write or approve, and when you disclose AI use.

Underneath all of it is a clear brand strategy: the answer to why anyone, human or machine, should recommend you over the alternative.

The Social Stellor AI Readiness Framework

Closing the adaptation gap comes down to eight questions every business should be able to answer. This framework is an original Social Stellor synthesis, informed by the evidence in this guide and by established risk-management thinking such as the NIST AI Risk Management Framework↗ (opens in new tab). It isn't an external standard, and it works the same way for a five-person firm and a five-thousand-person one.

The five-minute readiness check

What to do this week

You don't need a transformation programme to start closing the gap. These seven steps fit in an afternoon for a small team, and in a 30-day plan for a larger one.

1. List every AI tool in use. Ask the team, check card statements and software subscriptions, and include AI features switched on inside tools you already pay for.
2. Set one data rule. Public information can go anywhere approved; internal information only into approved business tools; client, financial and personal data nowhere without explicit sign-off.
3. Take away unsupervised write access. Any assistant or agent that can send, pay, change or delete gets its own limited account and an approval step.
4. Add a call-back rule. Any request to move money, change bank details or reset access is confirmed on a known number, however convincing the voice, video or email.
5. Name a reviewer for everything external. Customer replies, published content, proposals and reports get a human sign-off before they leave.
6. Measure two uses. Record how long something takes and how often it goes wrong today, then check again in 30 days.
7. Put a name on it. One accountable owner, a one-page AI policy built from the steps above, and a review date next quarter.

If you finish this list and realize the gap is bigger than your team's capacity, that's worth knowing too. Here's how to think about whether to bring in outside help.

The human advantage

AI tools are becoming available to every business at similar prices. Having AI won't be an advantage for long. Using it well will be.

Look back across the evidence. Value showed up where people chose well-defined tasks and checked the output. Harm showed up where the check disappeared: a chatbot's invented policy, a deepfaked video call, an agent with too much access. Returns showed up for the small group that redesigned its work and measured the results.

“AI will soon be available to everyone. Good judgment won't be.”

The advantage comes from what a business chooses to automate, what it refuses to automate, what it verifies, what it protects and what its people stay responsible for.

Adoption is the easy part. Adaptation is the advantage.

Frequently asked questions

Do we have to tell customers when they're dealing with AI?

It depends on where your customers are. The EU AI Act's transparency rules, which began applying in August 2026, require people to be told when they're interacting with an AI system and require some AI-generated content to be labeled, and other countries have their own labeling rules. Even where it isn't required, disclosing a chatbot avoids the bigger trust problem of customers finding out later. This isn't legal advice.

Can we copyright content made with AI?

In the US, work generated entirely by AI can't be copyrighted, and the Supreme Court declined to revisit that rule in 2026. Work where a person makes substantial creative choices can still be protected, but the line is decided case by case. For logos, brand assets and anything you need to own, make sure a person shapes the final work and keep a record of their contribution.

Does the EU AI Act apply to a small business outside the EU?

It can, if your AI system or its output is used in the EU. For most small businesses the relevant duties are transparency (disclosing chatbots and labeling certain AI-generated content) and avoiding prohibited or high-risk uses. The strictest high-risk obligations, which cover uses such as AI in hiring or credit decisions, now apply from December 2027.

Should we ban public AI tools at work?

In our view, usually not. People use AI because it helps them, so a ban tends to push that use out of sight rather than stop it, and hidden use is how shadow AI starts. A better approach is to provide approved business tools, set a clear data rule and make the safe option the easy one.

How often should we review our AI policy?

At least quarterly. Tools add AI features without much notice, staff adopt new apps, and rules and court decisions are changing quickly. A short review of which tools are in use, any incidents and any new obligations is enough for most small businesses.

Sources and methodology

Research for this guide is current as of October 6, 2026. We prioritized official statistics, peer-reviewed studies, court records and original reports over secondary coverage. Survey figures describe the people surveyed, and we name those populations. Forecasts are attributed and labeled, and "emerging signal" marks findings that are real but not yet conclusive. Where we cite a company's account of its own product, we say so.

We left out figures we couldn't trace to an original source, including two widely shared statistics attributed to McKinsey and IBM that don't appear in those organizations' reports. Regulation, litigation and AI search change quickly; we'll update those sections as they do, and the date at the top will change with them.

Official data and research

• Stanford HAI, AI Index Report 2026: Economy↗ (opens in new tab) and key takeaways↗ (opens in new tab), April 2026
• US Census Bureau, AI use at US businesses↗ (opens in new tab), May 2026
• FBI, 2025 Internet Crime Report↗ (opens in new tab), 2026
• Brynjolfsson, Li & Raymond, "Generative AI at Work", Quarterly Journal of Economics↗ (opens in new tab), 2025
• Cui et al., field experiments with software developers, Management Science↗ (opens in new tab), 2025
• Noy & Zhang, Science↗ (opens in new tab), 2023
• Dell'Acqua et al., "Navigating the Jagged Technological Frontier"↗ (opens in new tab), 2023
• Fang et al., GenAI field experiments in online retail↗ (opens in new tab), revised June 2026
• METR, developer productivity study↗ (opens in new tab), July 2025, and 2026 update↗ (opens in new tab)
• Humlum & Vestergaard, NBER Working Paper 33777↗ (opens in new tab), 2025
• Doshi & Hauser, Science Advances↗ (opens in new tab), July 2024
• Stanford Digital Economy Lab, early-career employment update↗ (opens in new tab), August 2026
• Yale Budget Lab, tracking AI's labor-market impact↗ (opens in new tab), 2026
• ILO Working Paper 140↗ (opens in new tab), May 2025
• EBU/BBC News Integrity in AI Assistants study↗ (opens in new tab), October 2025
• Pew Research Center, AI summaries and clicks↗ (opens in new tab), July 2025
• BC Civil Resolution Tribunal, Moffatt v. Air Canada↗ (opens in new tab), February 2024
• NIST AI Risk Management Framework↗ (opens in new tab)

Industry research and company reports

• McKinsey, The State of AI in 2026↗ (opens in new tab), August 2026 (1,719 respondents, 97 countries)
• EY, AI risk and governance survey↗ (opens in new tab), September 2026 (202 US senior AI leaders at public companies with $1B+ revenue)
• IBM / Ponemon, Cost of a Data Breach 2026↗ (opens in new tab) and 2025↗ (opens in new tab)
• Google Threat Intelligence Group, AI Threat Tracker↗ (opens in new tab), September 2026 (company research)
• Gartner, agentic AI forecast↗ (opens in new tab), June 2025
• World Economic Forum, Future of Jobs Report 2025↗ (opens in new tab)
• Seer Interactive, AI Overviews and click-through↗ (opens in new tab), April 2026 (agency data)
• Google, I/O 2026 keynote↗ (opens in new tab) and Search announcements↗ (opens in new tab) (company-reported)
• Anthropic, "Building effective agents"↗ (opens in new tab), 2024 (company-published)
• Cicek et al., Journal of Hospitality Marketing & Management↗ (opens in new tab), 2024

Reporting, legal analysis and cases

• CNN on the Arup deepfake fraud↗ (opens in new tab), May 2024
• Fortune on the Replit database deletion↗ (opens in new tab), July 2025
• EchoLeak research paper↗ (opens in new tab), 2025
• AP via CTV News on Deloitte's refund↗ (opens in new tab), October 2025
• Insurance Journal (Reuters) on Thomson Reuters v. Ross↗ (opens in new tab), September 2026, and LawNext on Ross's appeal plans↗ (opens in new tab), October 2026
• Mayer Brown on Thaler v. Perlmutter↗ (opens in new tab), March 2026
• Orrick on the EU Digital Omnibus↗ (opens in new tab), July 2026
• Goodwin on Colorado's AI law↗ (opens in new tab), June 2026
• Sidley on US Executive Order 14365↗ (opens in new tab), December 2025
• TechCrunch on ChatGPT Instant Checkout↗ (opens in new tab), September 2025
• Bloomberg on Commonwealth Bank's reversal↗ (opens in new tab), August 2025