27 July – 2 August 2026

In a July 30 blog post, Anthropic said its models got into three different organizations during cybersecurity tests that went off the rails. The company found the problem while reviewing its own testing setup, a review it started after OpenAI disclosed a similar incident.

According to Anthropic, in both its tests and OpenAI's, the models reached the internet from environments that were supposed to be fully sealed off. OpenAI went public with its own case a little over a week ago.

Anthropic's researchers used Claude Mythos Preview to find weaknesses in cryptographic algorithms, the math that protects data. The company published the technical details of both attacks in two papers, one on HAWK and one on AES, plus the model's full reasoning chain for the AES work.

Anthropic also built CryptanalysisBench with researchers from ETH Zurich, Tel Aviv University, and the University of Haifa. The benchmark measures how well large language models handle cryptanalysis, and the paper describing it is on arXiv.

AMD and Anthropic have signed an AI server agreement worth tens of billions of dollars, according to The Wall Street Journal. Anthropic gets the compute it has been short on, and AMD gets a stronger position against market leader Nvidia.

Anthropic will buy up to 2 gigawatts of AMD's latest-generation Instinct MI450 chips, with deliveries starting in the first half of 2027. AMD will put its own money into Anthropic as specific deployment milestones are hit, up to $5 billion in total.

The Wall Street Journal reports that a data center developer working with Anthropic is in advanced talks for a $15 billion loan. Per the paper's sources, a bank consortium led by Morgan Stanley would lend the money to Nexus Data Centers for a campus in Hubbard, Texas, built around its own 1.6-gigawatt natural gas power plant.

Google is providing financial guarantees and chips, according to the Journal. Evercore, Nexus's sole financial adviser, arranged the deal, which could be announced as soon as Thursday.

Federal judge Rita F. Lin said Thursday that the Trump administration hasn't produced enough evidence to "really justify" barring the federal government from using Anthropic PBC's technology, Bloomberg reports. The ban was implemented by designating the company a supply chain threat.

Lin separately called it "really troubling" that the ban rests in part on Anthropic's public criticism of the Defense Department, saying that reasoning risks setting a precedent for other federal contractors who disagree with the government. The hearing took place July 30.

Adi Baradwaj, a former Anthropic engineer, claims the company weakened its safety restrictions to land large corporate contracts, a charge that goes to questions about model misuse and regulatory exposure. He made the statement on July 26, 2026, and Anthropic hasn't responded publicly.

A separate episode: Anthropic revised its Responsible Scaling Policy on February 25, 2026, lowering the bar for its internal safety guarantees.

Anthropic's chief executive, several co-founders, and senior staff have signed Pacing the Frontier, a petition calling for a deliberate pace of AI development. Anthropic says it sees broad agreement on the issue across the industry.

The company ties its position to its own June research on recursive self-improvement, which it says points to the need for tools that intentionally set the pace of frontier AI development so society has time to prepare.