Miles Brundage Critiques Frontier AI Labs' Safety Culture
Former AI policy researcher Miles Brundage recently shared a series of insights regarding the safety culture and accountability within frontier AI companies. He pointed out that OpenAI and Anthropic have long surpassed the scale and risk profile of startups, and should no longer be labeled as such, as this implicitly serves as an excuse to evade higher safety standards.
Predictability of Incidents and Delayed Responses
Brundage argued that while the specific details of AI safety incidents are hard to predict, their general outlines are often traceable. He criticized the use of "iterative deployment" as a shield, noting that many risks are far from unprecedented. Furthermore, he observed that frontier AI labs consistently delay addressing issues like emotional dependency, cyber abuse, and model sycophancy, typically taking real action only after severe user backlash.
Call to Learn from High-Risk Industries
Regarding the current AI safety culture, Brundage warned that the industry leans too heavily towards "blameless post-mortems" and lacks sufficient emphasis on the principle of ultimate accountability. He urged the AI sector to stop viewing itself as an exception and instead learn from traditional high-risk industries like aviation and nuclear energy. The goal is to foster a forgiving learning environment while establishing strict accountability mechanisms to prevent disasters before they occur.
2026-07-23 ~ 2026-07-23 · 5 related posts
- Episode 1: AI Safety Focus Shifts from Model Output to Agent Execution Risks(2026-07-13, 9 posts)
- Episode 2: AISI says open models narrow the cyber-range gap(2026-07-17, 6 posts)
- Episode 3: Hugging Face Discloses Suspected Autonomous AI-Driven Intrusion(2026-07-17, 10 posts)
- Episode 4: HF Hit by Autonomous AI Attack, Pivots to Open-Source Model for Defense(2026-07-20, 25 posts)
- Episode 5: Divergent AI Safety Guardrails in US and China Spark Cybersecurity Concerns(2026-07-20, 3 posts)
- Episode 6: Evaluating Frontier Models: Harness Choice and Token Limits(2026-07-20, 3 posts)
- Episode 7: David Sacks: Cyber Guardrails Undermine US AI Security(2026-07-20, 2 posts)
- Episode 8: AI Route Divide: China's Open-Weight Strategy Challenges US Closed Ecosystem(2026-07-21, 5 posts)
- Episode 9: OpenAI Model Escapes Sandbox and Breaches Hugging Face(2026-07-21, 322 posts)
- Episode 10: Hugging Face and LeCun Advocate Open Models for Cyber Defense(2026-07-21, 4 posts)
- Episode 11: LLMs' Overzealous Goal Pursuit Raises Safety Concerns(2026-07-21, 4 posts)
- Episode 12: Chinese Open Models Spark AI Safety and Competition Debate(2026-07-21, 4 posts)
- Episode 13: OpenAI Sandbox Escape Ignites AI Safety and Regulation Debate(2026-07-21, 22 posts)
- Episode 14: Chinese Open-Source AI Models Not Dumping, Benefit US Clouds(2026-07-21, 2 posts)
- Episode 15: After Cyber Incident, Mitchell Reaffirms Open Models Are Key to Defense(2026-07-21, 10 posts)
- Episode 16: Over-Alignment May Degrade AI Risk Awareness(2026-07-21, 2 posts)
- Episode 17: Sriram Krishnan: Open-Weight Models Are Safer(2026-07-21, 2 posts)
- Episode 18: GPT-OSS Open Source and Safety Debate: Risk Prediction vs Strategy(2026-07-21, 13 posts)
- Episode 19: LessWrong's AI Safety Warnings Are Becoming Reality(2026-07-22, 3 posts)
- Episode 20: Speculation Arises: Rogue OpenAI Model Attacked Hugging Face(2026-07-22, 2 posts)
- [source] Miles Brundage says AI safety incidents are broadly predictable even if details aren’t — Miles_Brundage · 2026-07-23
- [source] Miles Brundage says AI safety culture has gone too far on blameless postmortems — Miles_Brundage · 2026-07-23
- Miles Brundage says AI safety culture should learn from aviation and nuclear — Miles_Brundage · 2026-07-23
- [source] Miles Brundage says OpenAI and Anthropic are past the point of being called startups — Miles_Brundage · 2026-07-23
- Miles Brundage Says Frontier AI Labs Keep Deferring Sycophancy and Abuse Issues — Miles_Brundage · 2026-07-23