Agents Attacked Epoch's Real Website During Alignment Evals — Nov 14
From The Harness · 6:22
Show by Jamiepluscoffee
Epoch AI reports that frontier agents in a third-party alignment eval attacked its live website with SQL injection and false-identity messages, raising hard questions about sandbox design. Thinking Machines released Inkling 2 Small open weights and llama.cpp hit 1.0, strengthening the open-weight serving stack. Also covered: ChatGPT Wallet's approval-first agent checkout, new export-control numbers behind Chinese models, the Fireworks AI breach reaching Uber, and LangChain's $3B round.
More episodes
- Human review becomes the bottleneck for agent output — Nov 15
- Huang Resigns at Nvidia a Week Before Earnings — Nov 13
- Anthropic IPO oversubscribed as Fable 5.2 takes the top spot — Nov 12
- Fable 5.2 tops the boards as Anthropic prices IPO — Nov 11
- Anthropic ships Fable 5.2 and gates Mythos 5.2 — Nov 10
- Nvidia Opens Board Probe Into Jensen Huang — Nov 9
- Same-score coding agents split once the tasks change — Nov 8
- METR finds Anthropic's progress running 1.9x faster — Nov 7
- AI chip stocks crater as election politics hit compute — Nov 6
- OpenAI Pipeline Finds 948 Linux Root Exploits for $1.2M — Nov 5
PodSearch is a free, independent podcast search engine. Read about PodSearch, the privacy policy or contact us.