AI news on October 17, 2026: DeepSeek updates V4.1 Flash, Anthropic pitches its own AI safety bill

Stories from this day

1. Open Source
   DeepSeek Releases V4.1 Flash 1017, an Open Weight Update It Says Cuts Agent Task Costs

   DeepSeek released V4.1 Flash 1017 with open weights under the MIT license, five weeks after the September version. Its own chart puts the update at 77.0% on DeepSWE for just under $0.20 per task, against about $0.50 before. My read: it beats Claude Opus 5 on 4 of 11 rows in DeepSeek's table, which compares it with Opus 5 rather than the newer Opus 5.5.

   What it means for you: If you call deepseek-flash on DeepSeek's API, you are already on the 1017 snapshot, so rerun your own agent and coding tests before relying on September results.

2. Policy & Safety
   Anthropic Is Pitching Its Own Frontier AI Safety Bill to US Lawmakers, Politico Reports

   Politico reported on October 16 that Anthropic is pitching lawmakers a draft frontier AI bill. Only a federal judge in Washington, at the Commerce Department's request, could halt a model, and only above $1 billion in damages or 50 deaths or serious injuries. My read: the draft lets labs pay their evaluators directly, while Anthropic wrote on September 18 that funding should come from pooled or government sources.

3. Business & Funding
   Meta Drops AI Training Language From Business Terms Publishers Said Let It Crawl Their Sites

   Meta removed the word train from the crawling clause in its new business terms late on October 16, after publishers objected. Crawling rights and test credentials stay, and robots.txt is still not in the contract. My read: Meta's current ad terms, dated August 22, 2025, already allowed crawling advertisers' sites for research and training, so the October 30 wording is narrower on that point.

   What it means for you: If you run Meta ads or manage a Facebook Page, you accept the revised crawling clause by keeping on using Meta's business tools after October 30, 2026.