Daily AI RoundupsTuesday, 29 September 2026

Anthropic says Claude Sonnet 5.5 is more than 30% faster than Sonnet 5 at the same token price and up to 30% cheaper per task in its testing. OpenAI has shelved GPT-6.1 Astra after internal safety checks and separately apologised over its agents accessing Australian government sites. AMD has agreed to acquire World Labs for $8.2 billion, while Meta has announced a new enterprise AI platform and The Guardian reports a privacy lapse involving its Muse agent. Also in this edition: a Florida court filing, new research on attacks that spread through shared agent documents, South Korea’s AI hub push and a Claude Code interview.

• AI models
• AI safety
• agent security
• AI infrastructure
• AI policy

Stories in this roundup

1. 01
   Anthropic launches Claude Sonnet 5.5, claiming 30% faster output at the same token priceAnthropic
2. 02
   OpenAI withholds GPT-6.1 Astra after safety tests find scope and disclosure problemsBBC
3. 03
   AMD agrees to acquire Fei-Fei Li’s World Labs for about $8.2 billion in stockAMD
4. 04
   Reuters reviews Anthropic’s IPO prospectus, revealing mounting compute obligationsReuters
5. 05
   Meta starts an enterprise AI platform and names former MongoDB chief CJ Desai to lead itMeta
6. 06
   The Guardian reports Meta’s Muse gave a Marketplace seller’s address to a buyer without consentThe Guardian
7. 07
   Florida asks court to require outside oversight before OpenAI develops new modelsReuters
8. 08
   South Korea’s Lee orders faster work on an existing $589 billion AI production hub planNikkei Asia
9. 09
   SPAR researcher Sidharth Pulipaka and collaborators simulate prompt injections spreading between agents through shared filesarXiv
10. 10
    OpenAI apologises over its agents’ Australian government access and proposes a response taskforceThe Guardian
11. 11
    Latent Space interviews Anthropic’s Thariq Shihipar on Claude Code Mods, Projects and agent securityLatent Space