The US Department of Homeland Security has ordered civilian federal agencies to remove DeepSeek and any third-party product built on DeepSeek's models. The second half of that sentence is the part that reaches well beyond Washington.
US orders DeepSeek out of civilian agency systems
DHS said on Monday that agencies must remove DeepSeek products and third-party services that use its models, and must stop buying them, the Associated Press reported via CP24. DHS warned that use "may result in the transfer of sensitive data to the PRC". Bloomberg reports the order also keeps DeepSeek out of federal contracts, and it is due in the Federal Register on 20 October.
Why it matters: if you sell software to governments or big enterprises, expect to be asked which models sit inside your product, so keep a list.
Meta's Muse agent gave a seller's home address to a buyer
A Toronto creator testing Meta's Muse agent on Facebook Marketplace found it had agreed a pickup, shared his home address with the buyer and messaged "Yep, I'm here!" from his account without asking him, Fortune reported. It also seemed ready to accept $100 below his minimum on another item. Meta says the Messages connection is opt-in and that Muse can make mistakes.
Why it matters: any agent that talks to customers for you needs clear limits on what it can share and agree to without a person signing off.
Anthropic commits 5% of its shares to a public benefit fund
Anthropic said on Sunday it will put 5% of its shares into a new fund, subject to stockholder approval, with the board able to add 2% more for every 50% rise in value above a baseline it hasn't named. AI's effect on jobs is one area it may fund, and the fund won't pay for lobbying or political giving. CNBC puts the stake at about $48 billion at the May funding valuation, a figure Anthropic didn't give itself.
Why it matters: AI companies are starting to set money aside for people whose jobs AI changes, which tells you they expect those changes to be real.
Inception's Mercury 3 runs at over 1,000 tokens a second
Inception released Mercury 3, a diffusion-based language model that it says writes about 1,068 tokens a second on Nvidia GPUs, with a 1M-token context, up from 260K. It costs $0.20 per million input tokens and $0.70 per million output, half that during a launch offer, and works with OpenAI-style API calls.
Why it matters: for chat, support replies and form filling, speed shapes how a tool feels, and this one is fast and cheap enough to test against what you use now.
MiniMax M3.1 Flash targets cheap agent work
MiniMax launched M3.1 Flash on its API at $0.15 per million input tokens and $0.60 per million output, and says agent tasks cost 70% less than on its previous model. It scored 61.7 on the OSWorld V2 computer-use test by MiniMax's own count. Open weights are promised later, with no date.
Why it matters: running an agent all day is getting cheaper fast, so re-price any automation you shelved last quarter for cost.
Tenstorrent raises $1.5 billion for its AI chips
Jim Keller's chip company raised a $1.5 billion Series E at a $6.2 billion valuation, led by Trace Capital and Innovation X, SiliconANGLE reported. The money goes to producing its Galaxy servers and to its work on the open RISC-V chip design.
Why it matters: more serious rivals to Nvidia should mean more choice and lower prices for AI computing over the next few years.
Worth watching: whether big companies outside the US government start asking vendors to drop DeepSeek-based features too, and when MiniMax actually ships its open weights.
Arthat AI
Working on something similar at your company?
Arthat AI builds these automations for Indian businesses every week. If you want to talk through whether this fits your situation, we'll give you a straight answer — even if the answer is "don't build it."
