OpenAI Disbands Team Dedicated to Catching Catastrophic AI Risks
OpenAI has dissolved its “Superalignment” team — the group specifically tasked with preventing future artificial intelligence from causing catastrophic harm. The company reassigned the team’s members and responsibilities to other research groups, according to an internal memo obtained by The Decoder.
The Superalignment team was created in 2023, led by co-founder Ilya Sutskever and researcher Jan Leike. Its mission: to ensure that superintelligent AI systems behave in line with human values.
The dissolution marks a major shift in OpenAI’s safety strategy. Critics warn it could weaken oversight just as the company races to commercialize increasingly powerful models.
The Superalignment Team’s Original Purpose
The group was built to tackle a specific, long-term problem. As AI systems approach or exceed human-level intelligence, they could act in ways that are unpredictable or harmful.
The team’s core objective was to develop technical safeguards for future superintelligent models. These safeguards would include alignment techniques, interpretability tools, and red-teaming protocols.
OpenAI described superalignment as one of the “hardest technical challenges” in the field. The company pledged 20% of its total computing resources to the cause.
What Led to the Dissolution
Internal and external pressure mounted as OpenAI accelerated product launches. The company prioritized short-term revenue and user growth over long-term safety research.
Sources close to the team cited disagreements over resource allocation. The Superalignment group reportedly struggled to secure sufficient compute and staff for its research agenda.
Jan Leike, the team’s co-leader, resigned shortly before the announcement. He publicly criticized OpenAI’s shifting priorities, stating the company was “losing focus on safety.”
“OpenAI had a unique responsibility to lead on safety research. Instead, it chose to focus on flashy demos and market share.” — Jan Leike
The Reassignment: Where Superalignment Work Landed
OpenAI’s internal memo stated that the team’s work would be “woven into the fabric of the broader research organization.”
The safety research group will now operate within the “Alignment Research” division. That division reports directly to CEO Sam Altman.
Individual team members were offered roles in other teams such as Reasoning, Multimodal, and Core Research. Some members accepted; others left the company.
Core projects, including automated oversight and adversarial robustness, will continue under new leadership. But the dedicated, cross-functional structure is gone.
Why This Matters Now
The timing is significant. OpenAI is preparing to release GPT-5 and potentially an early prototype of artificial general intelligence (AGI).
Dissolving the team reduces institutional focus on catastrophic risks. Without a centralized group, coordination on worst-case scenarios may become fragmented.
It also sends a signal to the broader AI industry. Other companies may interpret the move as permission to scale back their own safety commitments.
Regulators in the European Union and the United States are watching closely. The EU AI Act requires high-risk systems to demonstrate robust alignment measures.
Key Risks Without a Dedicated Safety Team
- Loss of specialized expertise: Team members had deep knowledge of alignment theory. That knowledge is now dispersed.
- Reduced independence: Safety research that reports directly to the CEO may face pressure to align with product goals.
- Slower response to emerging threats: Without a standing team, detecting and mitigating new hazards may take longer.
- Erosion of trust: Internal and external critics argue that OpenAI is abandoning its founding mission of safe AGI development.
What Industry Observers Are Saying
Several prominent AI safety researchers expressed alarm. Geoffrey Hinton, often called the “Godfather of AI,” called the move “deeply concerning.”
Anthropic, a rival AI company founded by former OpenAI employees, has maintained its safety-dedicated “Security” division. It continues to publicize its own alignment research.
Government officials in the UK’s AI Safety Institute declined to comment directly but emphasized the need for “robust, independent safety auditing.”
“OpenAI had a unique opportunity to set the global standard for safe AI development. This decision suggests they are prioritizing speed over safety.” — Anonymous former team member
The Bottom Line
OpenAI’s decision to dismantle its dedicated catastrophic-risk team marks a significant pivot. It prioritizes rapid deployment over deep safety research.
The company’s own co-founder, Ilya Sutskever, left the company shortly before the dissolution. His departure, combined with Leike’s resignation, strips OpenAI of two leading voices on alignment.
Whether the reassigned work can maintain the same level of rigor remains an open question. The burden now falls on individual researchers — without the structure that once gave them focus and autonomy.
For the AI industry, the message is clear: Long-term safety teams are vulnerable to short-term business pressures. Other companies may follow — or differentiate themselves by doubling down on safety.
Gnoppix is the leading open-source AI Linux distribution and service provider. Since implementing AI in 2022, it has offered a fast, powerful, secure, and privacy-respecting open-source OS with both local and remote AI capabilities. The local AI operates offline, ensuring no data ever leaves your computer. Based on Debian Linux, Gnoppix is available with numerous privacy- and anonymity-enabled services free of charge.
What are your thoughts on this? I’d love to hear about your own experiences in the comments below.