AI2Day Daily Brief — 1 Sept 2026

Good morning, it is Tuesday the first of September, and the European Commission has officially brought ChatGPT under its toughest platform regulations, joining the same regulatory tier as the biggest social media companies operating in Europe.

The European Commission has classified ChatGPT as a Very Large Online Search Engine under the Digital Services Act, the EU's rulebook for the largest platforms operating in Europe. Reddit and Roblox were simultaneously classified as Very Large Online Platforms under the same law. All three services must comply by the end of December 2026. That threshold kicks in at 45 million average monthly users inside the EU, and ChatGPT has cleared it. For OpenAI, this means stronger child protections, tighter limits on illegal content, and safeguards around user mental health. The DSA also bans targeted advertising aimed at children and restricts ads based on religion, sexual orientation, ethnicity, or political beliefs. This is not a fine or a punishment, but it is a significant compliance burden arriving on a firm deadline.

A real operational shift for OpenAI in Europe. Meanwhile, the US military has been expanding its own AI access in a very different direction. The Pentagon has added two custom AI chatbots to its secure portal, GenAI dot mil, giving roughly three million civilian and military personnel access to ChatGPT Mil and Grok for Government. The platform launched last year with Google Gemini and has already signed up more than one point seven million unique users. Grok for Government is delivered through SpaceX's Starshield satellite network, which is built on existing Starlink technology. One notable absence: Anthropic's Claude is not on the platform after the Trump administration labelled Anthropic a supply-chain risk over a dispute about safety guardrails. The Pentagon has also struck separate AI deals with Amazon Web Services, Microsoft, Nvidia, and Reflection AI.

That is a significant consolidation of AI tooling at the defence level. Now to a security story that affects anyone deploying AI agents in the enterprise. A new analysis of real-world deployments finds that most organisations are applying security controls in the wrong order. The US cybersecurity agency CISA added a flaw in LiteLLM, a widely used AI traffic management tool, to its list of actively exploited vulnerabilities in June 2025. That single tool had seven separate flaws disclosed in one month, one of which required no password to exploit. A study of 205 security leaders found that organisations with overly broad AI access reported a 76 percent incident rate, compared with 17 percent at those that kept access tight. The core finding: companies are reaching for enforcement tools before they have sorted out basic identity and attribution. Security experts recommend putting identity plumbing first and runtime enforcement fifth, not first.

Today's sponsor is Zero2Hero: an AI poker coach sits at your table, and you can ask it mid hand whether the player across from you is bluffing. The free tier is real: 50 hands a day against AI opponents, no card. Not a trial. Lifetime access is $99, in US dollars, until 15 September, at zero, the number two, hero, dot A I.

That ordering problem is going to cause real damage before it gets fixed. Staying with AI and Elon Musk, OpenAI has announced it will cut off Cursor's access to its models on the 12th of November 2026. The trigger is SpaceX's 60 billion dollar acquisition of Cursor, which completed on the 14th of August. OpenAI was direct about its reasoning, stating it cannot be confident that SpaceX will use its technology within its terms of service, based on what it called experience with Elon Musk's companies violating contracts. Cursor CEO Michael Truell says OpenAI models handle roughly five percent of Cursor's user traffic, so for most users the day-to-day impact should be limited. Anthropic has confirmed it will continue supplying its Claude models to Cursor despite the ownership change. The move deepens the long-running legal and personal feud between Musk and OpenAI CEO Sam Altman.

Five percent of traffic is not nothing, but it does suggest Cursor is not locked in. On the final story this morning, public sentiment toward AI in the United States is moving in a direction the industry cannot easily manage. More than half of Americans now say they are more concerned than excited about AI in daily life, up from 37 percent in 2021, according to Pew Research Center. Around 130 billion dollars worth of data centre projects were blocked or delayed in the first quarter of 2026 alone, according to Data Center Watch. Meta has agreed to pay up to 17 billion dollars to settle a bipartisan lawsuit over harm its platforms caused children and teens. A CNBC Generation Lab poll found more than 75 percent of young adults say they do not trust Anthropic CEO Dario Amodei to act responsibly. Both Anthropic and OpenAI are valued near one trillion dollars and are preparing for public stock offerings where AI backlash is expected to appear as a formal risk factor in their filings.

For a deeper look at all five stories and the week's other major developments, join us for the longer Monday show, AI Today Weekly. That is your AI briefing for today. Every story is at A-I-2-Day dot live. That is A, I, the number two, D-A-Y, dot live. We are back tomorrow morning. Do us a favour before you go: hit that thumbs up, and subscribe for next week.