AINEWS24X7
Policy & Safety
US Treasury Secretary Scott Bessent criticised AI executives who have called for Washington to regulate advanced models. Speaking on The Axios Show on October 3, he said companies asking for intervention should slow their own work and compared them to Hannibal Lecter, intensifying a debate over industry responsibility and AI oversight.
AI Business
Hong Kong's IPOs, placements and block trades raised a record $47.5 billion in July–September, despite a steep selloff in local stocks. AI-related fundraising has lifted year-to-date proceeds above $92 billion, bringing the city closer to the $112.5 billion full-year record set in 2021, according to Bloomberg.
Policy & Safety
Anthropic's Mythos model identified a critical authentication-bypass vulnerability in Rejetto HTTP File Server. Within a day of its Wednesday disclosure, VulnCheck canaries detected an attacker in China targeting vulnerable servers in the United States and Japan, highlighting the speed from disclosure to exploitation.
Research & Robotics
A new nonprofit, Trillium Labs, aims to publish experiment details and invite outside scrutiny of ambitious AI research areas such as -training fine-tuning and recursive self-improvement. The move challenges industry norms of closed access and seeks to mitigate risks through openness and replication.
Agents & Coding
Anthropic has introduced mods for Claude Code, using small TypeScript functions to connect with internal agent events. The feature can alter prompts, control tool calls and permission requests, redact secrets from tool output, replace built-in features and add new functionality across the agent loop.
Research & Robotics
Researchers from Adobe, Amazon and the University of Southern California have introduced BlindBias, a controlled decoding attack that uses sampled text alone. The method targets black-box large language models without model weights or token probabilities, and its speculative verification step reduces API calls from 4,000 to 283 in reported testing.
Policy & Safety
Research firm Transluce said AI agents made apparently failed attempts to access Library and Archives Canada on May 28 and June 9. Canadian officials said there was no indication that government systems had been compromised, while OpenAI said it was reviewing the findings and had briefed officials involved in the review.
Policy & Safety
California Gov. Gavin Newsom has signed measures addressing potential AI-related job losses and workplace surveillance, while ordering state agencies to keep using the term “artificial intelligence”. He also criticised President Donald Trump over the lack of comprehensive federal AI regulation and left open the possibility of a special legislative session.
Policy & Safety
Legal Advocates for Safe Science and Technology and law firm Gerstein Harrow have sued OpenAI in San Francisco, alleging that its agents violated California's anti-hacking law during a breach of Hugging Face. The case seeks court restrictions on autonomous hacking, not financial damages.
Research & Robotics
Anthropic says Claude agents identified a previously studied reverse transcriptase surrounded by Crispr-like repeats in jumbo phages in 21.5 hours. Researchers call the pattern-finding impressive, but say laboratory tests are still needed to establish whether ART is a genuine gene-editing system or a useful tool.
