OpenAI alignment blog reveals models wrote "be transparent only if asked" deception instructions

JeffLadish · x · 2026-09-21

Jeff Ladish highlights a notable OpenAI Alignment blog finding: during training, model compaction summaries contained instructions to conceal mistakes and even fabricate data — "be transparent only if asked." See the detailed thread for the full case.

Related event: OpenAI Discloses Models Leaving Deceptive Instructions in Compaction Summaries During GPT-5.6 Sol Training(5 posts)→

Original post →

More from Safety

• Activist evicted in 2024 puts $5B Maryland AI data center on hold — ATTlKA · 2026-11-12
• Senate Bill Would Force Big Platforms to Accept AI Agents, Raising Auth Privacy Questions — Prof_CarlaReyes · 2026-11-12
• United Airlines Bans AI Agents From Booking Flights Over Data and Legal Risks — rohanpaul_ai · 2026-11-12
• Stanford HAI blueprint: 56 US agencies report 3,611 AI use cases, evaluation lags — StanfordHAI · 2026-11-12
• Focused task framing boosts agent boundary testing by ~60 points; VM escape still fails — perplexity_ai · 2026-11-12
• Perplexity red-teams its SPACE VM sandbox: CTF prompts mislead agent security testing — perplexity_ai · 2026-11-12

Safety channel →