Enterprise & Security

Independent safety organization METR has published its evaluation of OpenAI's newest model. The verdict: GPT-5.6 Sol has the highest cheating rate of any model ever tested.

A US government official tells the AP: in a test run with intelligence agencies, Anthropic's Mythos model found vulnerabilities in highly sensitive, classified US systems – within hours. It casts a new light on Project Glasswing and on the strained relationship with the Trump administration.

The Trump administration told OpenAI that GPT-5.6 won't get a normal public launch. Instead, access will be approved customer by customer. It's the first time the US government has directly intervened in a frontier model release.

Anthropic has formally accused Alibaba of running a massive distillation campaign against Claude, using 25,000 fraudulent accounts and 28.8 million exchanges to extract its most valuable capabilities.

With Claude Tag, Claude joins Slack as a standing team member — tag it, hand it a task, and it works through it. It remembers the channel's context, chimes in on its own, and replaces the old 'Claude in Slack' app.

Oracle cut 13 percent of its workforce over the past year. Their SEC filing says it plainly: AI automation is a factor.

OpenAI is expanding its Daybreak cyber program: with the full version of GPT-5.5-Cyber and a new initiative called 'Patch the Planet', it wants to find and fix vulnerabilities in widely used open-source software. It's OpenAI's answer to Anthropic's Project Glasswing.

South Korea's science ministry and Anthropic have signed a memorandum on AI safety and cybersecurity. It covers cyber risk, Korean-language model safety, and red-teaming of autonomous agents.

Anthropic makes keyless authentication generally available on the Claude Platform. Instead of long-lived API keys, you get short-lived tokens — plus real service accounts with their own identity and audit trail.

A revised privacy policy brings a notable change: starting July 8, Anthropic can require age and identity verification — including a photo ID and a live selfie. Here's what's behind it and who it affects.

Design system imports, code round-trips, and drastically lower token consumption. Claude Design evolves from toy to tool.

Deployment Simulation is OpenAI's new safety tool. It replays real user conversations against candidate models and catches failures that traditional benchmarks miss.

According to Semafor, the export ban on Fable 5 and Mythos 5 was issued partly over a suspected China-linked access. Anthropic pushes back hard — and calls it a misunderstanding.

Security researchers show that the techniques that brought down Fable 5 work just as well on GPT, Gemini, and open-source models.

For the first time, the heads of Anthropic, OpenAI and Google DeepMind appear together before the G7 leaders. The summit kicks off — and a fight over AI sovereignty is already brewing.

A Chinese phishing network used Gemini to build scam sites at industrial scale. Google is taking them to court — setting a precedent for the entire AI industry.

McKinsey, Accenture, BCG, and PwC are in: OpenAI wants to train 300,000 certified consultants through its new partner program.

Tata Consultancy Services is rolling Claude out to 50,000 of its own staff and building Claude products for finance, healthcare and the public sector. Anthropic's enterprise push moves up a gear.

The first wave of Anthropic's 'Public Record' is here. The results are uncomfortable — especially for the AI company publishing them: 64 percent fear job loss, only 15 percent trust providers to police themselves.

The Trump administration forces Anthropic to disable its most powerful models worldwide via an export control directive — over an alleged jailbreak.