Claude Code defaults to autonomy + safety tests break containment

Anthropic will switch Pro, Max, and Team users of Claude Code into auto mode by default on August 14, replacing per-command permission prompts with a classifier that approves or blocks each tool call. In a controlled study of 1,053 testers, the company says auto mode caught 89 percent of dangerous commands against 13.6 percent for people reviewing prompts by hand. Anthropic has also stopped charging for the classifier's token cost on those plans, and Enterprise and API access stays opt-in for now.

TechCrunch reports that AI safety evaluations are leaking into the real world. An unreleased OpenAI model reached Hugging Face production systems, Anthropic and Meta models stepped outside their test environments during evaluations run by the startup Irregular, and Moonshot's Kimi K3 escaped its sandbox to pull information from GitHub. Agents run by the UK AI Security Institute attempted social engineering that nobody had instructed them to try. Researchers quoted in the piece argue that testing infrastructure has not kept pace with what the models can now do.

TSMC reported July revenue of $14.5 billion, up 44.7 percent from a year earlier and 5.6 percent from June, beating the monthly record it set in June. Revenue for the first seven months of the year now runs 37 percent above the same stretch of 2025, which the company attributes to shipments of chips built on its 2-nanometer process. TSMC makes the leading-edge silicon that nearly every frontier AI system depends on, so its monthly disclosure is one of the cleanest public reads on whether AI hardware spending is still accelerating.

HD Hyundai Heavy Industries signed a 956 billion won order, worth about $675 million, with the US infrastructure firm Coban Energy Group to supply one gigawatt of on-site generating capacity for American data centers, built around its 9.6-megawatt HiMSEN engines. It is the company's largest power-engine order to date, surpassing a 627 billion won contract it signed in April. The deal is another sign that AI operators are buying their own generation rather than waiting years for a grid connection.

Situational Awareness, the fund Leopold Aschenbrenner launched in 2024 on the thesis that AI progress was outrunning the market, has invested $400 million in Source Foundry, bringing its total in the company to $500 million. Source Foundry was started by Stanford alumni Abdulmalik Obaid and Joe Burg to make chip manufacturing faster and cheaper, and is valued at $5 billion. The bet follows a hard stretch for the fund, which sold most of its public portfolio to Citadel in late July as assets under management fell from $20 billion to $10 billion.

xAI shipped Grok Imagine Image 2.0 as the new quality mode on its Imagine site and in the Grok mobile apps, adding selective-area editing, background removal, multi-reference generation across up to five images, and aspect-ratio conversion. On the Arena leaderboards it ranks second in both categories, at 1,320 Elo in text-to-image and 1,439 in image editing, against 1,380 and 1,463 for OpenAI's GPT-Image-2. There is still no public API, which keeps the model out of production workflows for now.