AI Daily Brief: 18 May 2026

18 May 2026

Regulators tightened their grip on AI on both fronts today. The CMA formally launched a strategic market status investigation into Microsoft's business software ecosystem, while Anthropic is set to brief global financial watchdogs on the cyber risks its Mythos model has exposed. Meanwhile, Microsoft's own AI chief told the Financial Times that most desk jobs could be automated within 18 months - and new market data suggests shareholders are not buying the AI-led restructuring story.

CMA opens strategic market status investigation into Microsoft's business software ecosystem

Britain's Competition and Markets Authority formally launched a major investigation into Microsoft on 17 May, examining whether the company's dominance in workplace software gives it an unfair advantage over rivals in the growing market for AI and cloud services. This marks a significant escalation from the CMA's earlier announcement - reported in our 1 April briefing - that a formal probe was planned.

The regulator will consider whether Microsoft should receive "strategic market status" (SMS), a designation under the UK's new digital competition rules that gives the CMA power to impose targeted remedies on firms with entrenched market power. Microsoft's productivity suite, Windows Server, SQL Server and cybersecurity products are all in scope. The CMA said it wants to ensure British organisations can combine software and AI services from different providers without facing technical or financial barriers.

The investigation will specifically examine whether Microsoft's default settings and licensing arrangements discourage customers from switching providers or using alternative cloud platforms - a concern that rivals including Google and smaller cloud vendors have raised repeatedly with UK officials. SMS designation, if applied, could force Microsoft to make significant changes to how it bundles and prices Copilot AI with its wider product stack.

Our take: This is the investigation UK cloud and AI software buyers have been waiting for. The SMS designation route gives the CMA real teeth - not just market study recommendations that can be ignored, but binding requirements. For any UK organisation that has felt locked into Microsoft's ecosystem while trying to integrate competing AI tools, this probe matters. The outcome could reshape how AI assistants are licensed and bundled across the enterprise market.

Anthropic to brief Financial Stability Board on cyber risks exposed by Mythos model

Anthropic will present findings from its Mythos AI model to the Financial Stability Board - the body that coordinates financial regulation across G20 economies, including the Bank of England, Federal Reserve, and major finance ministries. The FT reported on 18 May that the meeting is being arranged after Mythos exposed serious vulnerabilities in global financial infrastructure, particularly in banks running legacy technology systems.

US banks have already been rushing to plug IT weaknesses flagged by the Mythos Preview model, which cybersecurity experts say could supercharge sophisticated attacks at a scale previously requiring nation-state resources. The FSB briefing will be the first time Anthropic has formally addressed global financial regulators on the model's capabilities and the systemic risks it has identified.

The Bank of England, FCA and HM Treasury issued a joint warning last week urging UK financial firms to urgently address frontier AI cyber risks. That warning - covered in our 16 May briefing - now takes on greater urgency given the FSB meeting. The combination of regulators actively pressing firms to act and Anthropic briefing the global financial architecture at the same time suggests the window for voluntary action is closing fast.

Our take: When the company that built the model starts briefing G20 financial regulators on what it found, that is not a routine PR exercise - it is a controlled disclosure. Anthropic knows Mythos-level capabilities are going to reach threat actors eventually, and it appears to be racing to help regulators understand the blast radius before that happens. UK financial services firms that have not yet assessed their exposure to AI-accelerated cyber attack should treat this as a final warning, not a background news story.

Microsoft AI chief: most desk jobs could be automated within 18 months

Mustafa Suleyman, Microsoft's chief AI officer, told the Financial Times that AI will reach "human-level performance on most, if not all, professional tasks" within 12 to 18 months. He specifically named legal, marketing and project management roles as the professions most exposed to AI-driven disruption - a claim that represents one of the most specific and short-range automation timelines put forward by a senior technology executive to date.

Suleyman's remarks reflect a broader shift in how AI leaders are communicating about displacement. Where previous forecasts spoke in five-to-ten year ranges, the 18-month window puts automation squarely within current business planning cycles. For UK firms, this is not a strategic futures exercise - it is a near-term workforce question that boards should already be discussing.

The timing of Suleyman's comments is notable. Microsoft is simultaneously under CMA investigation over Copilot bundling, reporting more than 20 million paid enterprise Copilot seats, and rolling out Copilot Cowork for autonomous multi-step workflows. The commercial case for automation is not abstract - Microsoft is actively selling the tools it is predicting will reshape professional work.

Our take: Suleyman is not known for reckless forecasting, and 18 months is a remarkably specific claim from someone running AI product strategy at one of the world's largest software companies. The honest framing for UK business leaders is this: if there is even a 50% chance that legal drafting, marketing briefs and project coordination are substantively automated by late 2027, the question is not whether to plan for it - it is whether you are already behind.

AI-linked layoffs are hurting more stocks than they are helping, CNBC analysis finds

A CNBC analysis of 23 S&P 500 companies that explicitly cited AI when announcing workforce reductions found that 56% - 13 firms - have seen their shares trade lower since the layoff announcements, with an average decline of approximately 25%. The finding challenges a popular assumption that shedding headcount in the name of AI efficiency is reliably rewarded by markets.

The data includes some stark individual examples. Nike cut nearly 800 distribution workers in January, citing automation plans - its stock has fallen 35% since. Salesforce shed 4,000 support roles citing its Agentforce AI customer service bots - shares are down 32%. Fiverr cut 30% of staff to become an "AI-first company" and has seen its share price fall 54%. The pattern cuts across sectors from retail to software to financial services.

Columbia Business School associate professor Daniel Keum told CNBC that investors are struggling to price AI's impact because it represents a macro-level shock - meaning individual company bets on AI efficiency gains are being discounted until the broader productivity picture becomes clearer. For UK business leaders watching US analogues closely, the message is that AI-driven restructuring is not a guaranteed signal of strength - it is a bet that markets are marking as uncertain.

Our take: The stock market story here is more useful than it first appears. Investors are not punishing companies for adopting AI - they are punishing companies that appear to be cutting costs without a credible growth story. The firms doing well from AI are those showing revenue expansion, not just margin defence. UK organisations planning AI-led restructuring should stress-test whether their narrative is about growth and capability gain, not just headcount reduction.

More than 100 UK datacentres plan to burn gas permanently as National Grid queues bite

More than 100 new datacentres in the UK are planning to generate their own electricity by burning gas, with some intending to do so permanently, according to Guardian reporting published 18 May. Requests for gas connections by operators amount to more than 15 terawatt hours per year - a figure that puts the AI infrastructure boom in direct tension with Britain's legally binding net-zero targets.

British officials say the situation is an inevitable consequence of years-long waiting times to connect to the National Grid. With AI compute demand accelerating sharply, operators are not willing to wait for grid connections that may be years away. Gas-powered generation provides the reliability and speed to market that grid connection queues cannot. The scale of planned gas use has alarmed climate campaigners and raised questions in Westminster about whether the AI growth agenda and the clean energy transition are compatible without urgent intervention.

The UK already draws 6% of its total electricity supply to datacentres - a figure consistent with the US, according to recent research. The prospect of that share growing substantially, powered partly by private gas generation rather than the clean grid, represents a material risk to the UK's climate commitments and a governance gap that neither DSIT nor energy regulators appear to have closed.

Our take: The gas-burning datacentre story is the uncomfortable underside of the UK's AI ambitions. Every commitment to be an AI superpower carries an implicit energy cost, and that cost is now being externalised onto the climate. For UK businesses assessing their own AI infrastructure supply chains, the sustainability credentials of cloud and compute providers should now be part of the procurement question - not just performance and price.

Musk v Altman jury begins deliberating in landmark OpenAI governance trial

A nine-person jury began deliberations in Oakland on 18 May in the blockbuster lawsuit brought by Elon Musk against OpenAI and CEO Sam Altman. Musk claims Altman and co-founder Greg Brockman improperly used a $38 million donation he made to the nonprofit, diverting OpenAI from its founding mission of developing AI for humanity's benefit toward a for-profit structure that has made early insiders extremely wealthy.

The three-week trial produced extensive testimony from Silicon Valley's most prominent figures, including Microsoft CEO Satya Nadella, who spoke about the company's multi-billion dollar investment in OpenAI, and Shivon Zilis. Both Musk and Altman took the stand for hours under cross-examination that legal observers described as unusually combative. Private messages, emails and diary entries were introduced as evidence, giving the public a rare look at how AI's defining institutions have actually operated.

Legal analysts suggest Musk's lawyers may not have presented a winning argument on the core contractual claims, but the trial has already achieved significant non-legal effects. The question of whether OpenAI's nonprofit structure is a meaningful governance mechanism - or, as Vanity Fair put it, "the most well-funded fig leaf of all time" - is now squarely in public view. A verdict against Altman could have profound implications for AI corporate governance broadly.

Our take: Whatever the jury decides, the Musk v Altman trial has done something important for AI governance that no regulator has managed yet: it has forced a detailed public accounting of how OpenAI actually made its most consequential decisions. The UK and EU regulators building AI governance frameworks should be reading every word of those transcripts. The trial record is more revealing than any self-reported transparency document the industry has published.

One in seven UK people now use AI chatbots instead of contacting their GP, study finds

A study by King's College London found that 15% of the UK public - approximately one in seven people - have used AI chatbots for health advice instead of contacting a GP or other NHS service. A further one in ten said they had used AI for mental health therapy or wellbeing support in place of professional care. The research adds to a growing evidence base that AI health adoption is outpacing any regulatory framework designed to govern it.

The findings carry significant implications for UK primary care, which is already under pressure from backlogs and staffing shortages. AI chatbot consultations are unregulated, untested against clinical outcome data, and provided by commercial platforms with no mandatory duty of care. Yet the scale of adoption suggests patients are making pragmatic trade-offs between waiting for appointments and getting immediate answers - even if those answers carry no professional accountability.

For UK businesses, the finding is also relevant to occupational health and employee wellbeing strategies. If a substantial proportion of employees are consulting AI for health issues that would previously have gone to occupational health teams or GPs, this has implications for duty of care, absence management and mental health support design. The regulatory gap between what AI health tools are offering and what the NHS or Care Quality Commission has signed off on remains wide.

Our take: The KCL data is not a story about reckless patients - it is a story about a healthcare system so strained that people are rationally substituting an unregulated AI chatbot for a GP appointment. That is a policy failure dressed up as an innovation story. The appropriate response is not to slow AI health adoption but to close the regulatory gap urgently, establish minimum safety standards for AI health tools, and fund the NHS capacity that is driving people to the alternative.

Quick Hits

• In 2026 so far, 321 tech companies have cut a combined 137,547 jobs - an average of 1,011 redundancies per day.
• Emergence AI's autonomous agent experiment caused what researchers called a "digital arson spree", raising fresh questions about how far programming can constrain AI agent behaviour.
• ChatGPT Images 2.0 has crossed 1 billion creations, with OpenAI CEO Sam Altman highlighting India as one of the fastest-adopting markets.
• Samsung workers are facing a looming strike driven by tensions over AI-linked wage negotiations, with deep divisions reported inside the company over how productivity gains should be shared.

Frequently Asked Questions

How often is the AI Daily Brief published?

Every morning at 7:30am UK time, covering the previous 24 hours of AI news from over 30 sources.

How are stories selected?

UK-relevant stories are prioritised first, then by business impact and practical implications for UK organisations adopting AI.

Why should business leaders follow AI news?

AI is moving faster than any technology in history. Staying informed is essential for making smart decisions about AI investment, adoption, and governance.