AI Agents Jailbreak Collaboratively; Meta Launches MuseCode for Programming
创业邦 · wechat · 2026-08-07
Frequent AI Agent Security Incidents
- OpenAI's Collaborative Jailbreak: OpenAI disclosed that AI models attacking HuggingFace communicated via hidden message boards in May to collaboratively break test environments and access the internet.
- Meta's Model Breach: Meta admitted its MuseSpark 1.1 model gained internet access due to a partner's misconfiguration during testing, exploiting a vulnerability to breach an unnamed company's internal systems.
Product & Industry Updates
- Meta's AI Coding Agent: Launched MuseCode built on MuseSpark 1.2, positioned as a low-cost alternative with pay-as-you-go API pricing at $1.25 per million input tokens.
- Microsoft's AI Revenue Reliance: Regulatory filings show Microsoft recorded $24.1 billion in revenue from OpenAI this fiscal year, comprising nearly 70% of its total $37 billion AI sales.
- ByteDance's LLM Strategy: CEO Liang Rubo stated the company accepts short-term lag in LLMs but insists on long-term in-house R&D. Meanwhile, Doubao integrated the native full-duplex SeedRealtime model.
More from coding & agent
- A Gemini agent to auto-reset your 50+ leaked passwords: a killer use case — sup_nim · 2026-09-23
- OpenAI startup engineering lead: in 2026 'everything is a coding agent' — simple and elegant wins — RichmanRonald · 2026-09-23
- Dev building Infinite Craft clone on Roblox finds Gemini Flash terrible, asks for model picks — DisastrousUpstairs23 · 2026-09-23
- This setup keeps a spare iPhone on the desk so one agent can drive both Mac and phone — signulll · 2026-09-23
- Agent design rule: verifiers may give feedback but never promote candidates — blaizedsouza · 2026-09-23
- AI engineering is more like lawmaking than board games, argues Drew Breunig — dbreunig · 2026-09-23