Daily AI ·
Anthropic Distillation Heist, OpenAI Cap War, and Safety Exit Lead AI News
Anthropic told Congress that Alibaba-linked actors conducted the largest known AI distillation heist against Claude. OpenAI retaliated against Anthropic by dropping ChatGPT usage caps, while its safety head resigned. Industry watchers also saw GPT-5.6 Sol top a key benchmark, and the FLI Safety Index hand out poor grades to leading labs.
10 sources1 min read
Anthropic reveals massive AI distillation heist by Alibaba affiliates Anthropic told the Senate Banking Committee that 28.8 million queries were run against Claude through 25,000 fake accounts by Alibaba-linked operators to distill its capabilities. Separately, the FT reported OpenAI and Google sold AI services to Singapore affiliates of Pentagon-blacklisted Chinese firms, exposing a regulatory gap. 1234
OpenAI drops ChatGPT usage cap in direct competitive strike at Anthropic OpenAI lifted the usage cap for Plus, Business, and Pro plans and reset all accounts, announcing it crossed 6 million active users. The move was widely read as a timed strike against Anthropic's uncertain Fable model subscription window, compounded by improved GPT-5.6 Sol efficiency. 567
OpenAI safety systems head Johannes Heidecke resigns amid restructure Heidecke’s departure is the latest high-profile exit as the company merges its safety and research teams under VP Mia Glaese. Chief Research Officer Mark Chen cited increased demands on safety due to faster model training cadences. 8
GPT-5.6 Sol tops Design Arena leaderboard, beating Claude Fable 5 Greg Brockman hailed the result as a 'big milestone' as Sol achieved a 1353 Elo score, surpassing Claude Fable 5 at 1345. OpenAI claims Sol also leads in coding, reasoning, and safety benchmarks. 9
Future of Life Institute AI Safety Index gives leading labs poor marks Anthropic scored a C+, OpenAI a C, and Google DeepMind a C in the Summer 2026 edition, with xAI, DeepSeek, and Mistral receiving F grades. The report flagged broad declines in risk assessment and governance since winter. 10
Daily AI
The most important AI & large language model developments — sourced and cited every morning.