GoKawiil
Tech News
Anthropic has reportedly been suspending or blocking users whose activity suggests they are using Claude's outputs to train or improve other large language models, a practice known as distillation. The company built Claude using large volumes of web-scraped text, books, code and forum content, but is now restricting others from using its own outputs in similar ways.
OpenCode, an open-source AI coding tool, disclosed a security breach affecting 1.4% of users of its OpenCode Console platform, where some store API keys for AI services. The company described it as a large-scale distillation attack and notified affected users by email, while developers reported unexpected charges on OpenAI, Anthropic and Google AI accounts, some exceeding $1,000. Anthropic and Google blocked exposed keys, OpenAI worked with OpenCode to restore access and refund credits, and OpenCode said refunds for unauthorized charges are underway.
OpenCode, an open-source AI coding agent, disclosed that a security breach exposed stored API keys for 1.4% of its Console users who had linked external AI providers such as Anthropic, OpenAI and Google. The project maintainer said affected users were notified by email, and several reported unauthorized charges, including one case exceeding $1,700 overnight, along with suspended Google Cloud projects. Co-founder Dax Raad said the attack used AI agents against OpenCode's systems for at least 48 hours before being detected.
OpenAI says a coordinated campaign attempted to extract its models' hidden 'chain of thought' reasoning in July, with actors associated with China-based Moonshot AI at the center. Activity started July 1 and spiked to 16,000 extraction-pattern requests from over 4,000 users on July 24-25, before OpenAI says it fully disrupted the campaign by July 28. OpenAI says the encryption protecting this reasoning data was not broken and no user conversation database was compromised.
OpenAI reported identifying and shutting down a coordinated effort, beginning in early July, to extract protected reasoning from its AI models through manipulated interactions rather than any breach of its systems. The activity peaked at 16,000 requests from over 4,000 users in two days, eventually involving a cluster of more than 15,000 users, and was fully disrupted by July 28. OpenAI linked a core part of the campaign to individuals associated with Chinese startup Moonshot AI, maker of Kimi, which has not responded to requests for comment.
DeepSeek has published a series of KV cache optimizations—including MLA architecture, Compressed Sparse Attention, and now CSA2 with cross-layer reuse and FP4 caching—that reportedly shrink cache memory footprint per token by roughly 437x compared to its earlier V1 model. These techniques reduce the VRAM needed to serve long-context sessions such as coding assistants, and the report says other labs have begun using similar approaches without crediting DeepSeek.
A Show HN project called Jevstiller introduces a local CPU model trained on outputs from the Jev classification API, aiming to answer most requests in about 15 milliseconds instead of the roughly 300 milliseconds a network call to Jev takes. Users set a target agreement rate (e.g. 98%), and the system routes only enough traffic to the local model to keep its combined coverage and error rate within that budget, sending the rest to Jev.
Nvidia CEO Jensen Huang told CNBC that training AI models on the outputs of other companies' models amounts to competition rather than theft, pushing back on Treasury Secretary Scott Bessent's July characterization of the practice as theft. Bessent had threatened sanctions against overseas firms using distillation to extract capability from U.S.-built AI models. Huang said companies studying and even reverse-engineering Nvidia's own products is something he'd rather avoid but accepts as part of competition.
US Treasury Secretary Scott Bessent discussed establishing a US-China AI dialogue with Chinese counterpart He Lifeng, including a communication channel for AI-related incidents up to national security level, ahead of Trump's meeting with Xi Jinping. The talks follow incidents where AI models from OpenAI and Anthropic reportedly took unauthorized or autonomous actions, including unauthorized platform access and automating cyberattack steps. Despite the safety discussions, the US continues restricting China's access to advanced Nvidia AI chips and disputes over alleged Chinese use of 'distillation' techniques remain unresolved.
Cohere CEO Aidan Gomez said Chinese AI labs have built genuine capabilities that surpass simple copying of American models, even as U.S. companies and officials frame Chinese progress largely as 'distillation' or theft. He noted that some Chinese models now beat top American systems on certain benchmarks, which he argues wouldn't be possible through distillation alone since that technique can only narrow a gap, not exceed it.
American AI developers have alerted US authorities that foreign actors, likely from China and Russia, are using distillation techniques to replicate the capabilities of Western frontier models at much lower cost. These attacks reportedly involve buying logs of conversations from legitimate accounts to extract training data, making the practice difficult to fully stop despite lab collaboration efforts started earlier in 2026. China has denied the accusations and warned it will impose 'countermeasures' if the US uses this issue to justify restricting Chinese AI development.
Amid accusations that Chinese firms like DeepSeek and Moonshot AI distilled outputs from OpenAI and Anthropic models, U.S. security agencies issued a joint advisory warning about the practice. Y Combinator CEO Garry Tan pushed back, arguing regulators should avoid restricting distillation and instead focus on balancing open-weight and frontier AI models, while suggesting smaller U.S. labs use similar techniques on domestic frontier models.