AI News, Oct 21, 2026: US orders federal agencies to stop using DeepSeek AI

Governments move on AI safety and DeepSeek while Google, Mistral and ElevenLabs push new tools for voice, cars and chips.

1. 01
   Anthropic proposes AI safety bill with judge sign-off for blocking modelsTechCrunch

   Anthropic proposed a safety bill that would bar federal agencies from blocking a model's release without a judge's permission, even when it could cause mass casualties.

   Why it matters

   It could shape how and when the government can stop a risky AI model from being released to the public.
2. 02
   Google brings personalized Gemini to Android Auto and cars with Google built-inThe Verge AI

   Google is adding Personal Intelligence to Gemini in cars, letting it draw on Gmail, Calendar, Maps, Photos, YouTube and Search.

   Why it matters

   Your car's assistant can use your email, calendar and maps to answer questions, if you allow access to those Google apps.
3. 03
   Anthropic reportedly struggles to get enough cloud computing powerTechCrunch

   Anthropic is reportedly having trouble securing compute from cloud providers that now see it as a competitor, as its spending rises ahead of a planned November IPO.

   Why it matters

   Limited computing capacity could affect how fast Claude improves and how reliably it runs for the people and businesses using it.
4. 04
   Google releases tool to move PyTorch apps off Nvidia chipsTechCrunch

   Google released a developer tool letting PyTorch users move AI apps off Nvidia GPUs, following a similar move by Amazon.

   Why it matters

   Easier switching between chip makers could widen options and eventually lower the cost of running AI services.
5. 05
   ChatGPT adds the ability to lock individual chatsTechCrunch

   OpenAI added a ChatGPT feature that lets users lock chats so they stay private when a device is shared with friends or family.

   Why it matters

   You can hand your phone to someone without them seeing private or surprise-planning conversations.
6. 06
   Meta pauses Muse chat sharing after shared conversations appear on the webThe Verge AI

   Meta paused Muse's conversation-sharing feature after security researchers at Claroc found more than 6,800 shared chat links publicly available on the web.

   Why it matters

   If you used Muse's share links, your conversations may have been publicly findable, so it is worth checking what you shared.
7. 07
   Mistral launches Voxtral 3 for building real-time voice assistantsThe Decoder

   Mistral AI launched Voxtral 3, a voice agent toolkit combining speech-to-text, text-to-speech and an SDK, supporting more than 50 languages and tool use during live conversations.

   Why it matters

   Voice assistants in apps and customer service could get faster and more natural, with options to keep data in the EU.
8. 08
   US orders federal agencies to stop using DeepSeek AITechCrunch

   The US ordered federal civilian agencies to remove DeepSeek products and services built on its technology, citing national security and possible data transfer to China.

   Why it matters

   Government workers and contractors must drop DeepSeek tools, and it signals more scrutiny of Chinese AI products used in the US.