Zvi (Don't Worry About the Vase)·16 hours ago·
42
● 10 sources
Anthropic’s federal affairs lead replied to criticism of a leaked Anthropic draft federal AI safety bill, arguing the company addressed concerns and that the draft reflected earlier policy ideas. The response says the bill text was circulated with a small number of congressional offices in September. As a result, Anthropic says it will update its discussion language on preemption and continue dialogue on model blocking, risk thresholds, and safety incident reporting, while critics maintain the draft is weaker and too self-interested.
Anthropic pitched a draft U.S. AI safety bill that would let independent evaluators flag certain high-severity risks from an AI model’s creation or release and prompt courts to stop distribution. It defines the threshold for a potential block as scenarios where 50+ people could die or $1 billion+ in property damage could occur. The proposal would create a 72-hour reporting window to the Commerce Department and require a D.C. judge to halt releases rather than allowing federal agencies to do it on their own.
Meta removed the word “train” from its commercial terms of service for business pages after AI-related backlash over data scraping concerns. The updated terms take effect on October 30. As a result, Meta changes the wording but did not clearly notify users about the language change, leaving uncertainty ahead of the rollout.
Anthropic describes its mission to help society navigate the evolution of intelligence and pursue discovery through research, policy, and its Claude product. The page specifically references Claude as part of its work. This doesn’t announce a new event or change plans; it functions as a mission/overview statement.
Meta has been ordered to identify executives who authorized torrent downloading tied to copyright disputes in AI training cases. The discovery order was issued by Magistrate Judge Thomas Hixon, and it forces Meta to clearly name key decision-makers tied to access to shadow libraries like Anna’s Archive and LibGen. As a result, Meta’s discovery process may expand beyond staff to include a broader chain of supervisors and leadership involved in early AI training dataset work.
The latest open-artifacts issue highlighted new releases and regulatory moves affecting open-weight AI, including a China pre-approval proposal for frontier models shared internationally and a U.S. DHS procurement restriction covering DeepSeek. DeepSeek’s inference stack, DeepInfer, is estimated to deliver up to 90% gross margin on tokens sold using that stack. As a result, more efficient inference tooling gets shared publicly while open-model sharing faces tighter government controls and vendors, potentially shifting prices and margins across the open-model ecosystem.
Effective altruism is being described as a new mainstream driver of “AI doom” thinking in Silicon Valley, with its supporters positioned across the AI industry including Anthropic and the newly appointed AI safety watchdog METR. The piece points to METR as part of the reason the view has spread beyond the fringes. As a result, debates about AI risk are framed through competing apocalyptic belief systems rather than purely technical concerns.
The article argues that workplace writing has been used as proof of work for compliance, effort, and learning, but LLM-generated text undermines that evidence. It says this became possible in 2023 when LLMs could be used to generate plausible reports, status updates, and learning artifacts cheaply. Organizations will have to change by using harder-to-forge evidence, banning certain AI-generated writing, or tracking effort in other ways, while engineers shift toward producing writing for communication rather than as evidence.
Lithography exports—specifically DUV immersion lithography systems—are central leverage in shaping China’s ability to scale AI chipmaking and compute. China imported about 90 immersion lithography systems per year in 2024–25, nearly two-thirds of ASML’s immersion exports. Congress’s proposed MATCH Act would restrict DUV immersion exports and servicing to reduce that leverage loss and buy time for US AI compute advantage into the 2030s.
The U.S. Department of Homeland Security ordered all U.S. civilian federal agencies to remove DeepSeek products (and some third-party products built using DeepSeek) from their IT systems, blocking future purchases unless exceptions are justified. The ban applies with 60 days’ notice for exceptions and covers DeepSeek models such as DeepSeek V4 Flash, which has 284 billion parameters. Agencies must audit installed software and routing tools, but it remains unclear whether locally running DeepSeek models on Windows PCs are covered, making Windows and similar setups potentially affected.
Mayor Zohran Mamdani told New Yorkers at a People’s AI Assembly event that AI deployment is not inevitable and urged public pressure for accountability. The talk followed his order banning AI use in elementary and middle schools last month. His stance pushed cities toward regulating and restricting AI directly rather than treating it as unavoidable federal policy gaps.
Every AI story that matters,
in your inbox by 8am.
TLDRocket reads all relevant sources, removes duplicate coverage, and summarises the
day in two minutes. Follow companies and topics for alerts, or get the
briefing in Slack. Free, no spam, unsubscribe anytime.
Reading TLDRocket needs no cookies, and the readership counts we rely on come from
our own cookieless analytics. Google Analytics is the exception: it sets cookies and
reports to Google, so it stays switched off until you allow it. You can change your
mind any time from “Cookie settings” in the footer.
Strictly necessary
Session security and form protection (tldrocket-session,
XSRF-TOKEN, 2 hours). The site cannot work without them,
so they need no consent.
Always on
Google Analytics 4 (_ga,
_ga_<id>, up to 2 years). Measures which
stories and sections readers use. Google acts as a third-party processor and may
store the data outside the EU. No advertising, no profiling, no data sold.
