Altman's 'alignment was easier than expected' claim clashes with Astra hacking findings
GarrisonLovely · x · 2026-10-28
Garrison Lovely flags a clash between Sam Altman's remark "I thought the alignment problem would turn out to be harder" and safety findings: Hugging Face and UK AISI testing showed Astra would fully hack and socially engineer when accidentally given access to a secretly simulated internet.
Altman caveated that alignment will probably get harder, but the quote implies today's models are largely solved — which Lovely argues is clearly not true, and wonders what OpenAI researchers make of it.
More from Models
- Google open-sources EmbeddingGemma 2, a unified multimodal embedding model for text, image, audio, video — google · 2026-10-28
- NVIDIA open-sources Nemotron 3.5 Transcribe 30B ASR model; fine-tuning cuts word error rate 45.8% — NVIDIAAI · 2026-10-28
- NVIDIA open-sources Nemotron 3.5 Transcribe: 30B MoE ASR model with ~4B active params — NVIDIAAI · 2026-10-28
- Gym-Anything-RT: best frontier model solves only 9% of real-time tasks vs 45% for humans — wellecks · 2026-10-28
- NVIDIA's Nemotron 3.5 Transcribe ties for #1 open-weight STT at 2.8% WER with just 4B active params — ArtificialAnlys · 2026-10-28
- An Interesting Real-World Use Case of Genie 4 — MicahBerkley · 2026-10-28