A Freedom of Information Act (FOIA) lawsuit revealed that OpenAI's Pentagon contract initially included a 'minimal refusal rates' clause for AI, but both parties now deny this language was finalized. The contract, worth up to $200 million, aims to develop AI tools for military use.
TL;DR
- OpenAI's Pentagon contract initially included a 'minimal refusal rates' clause, but both parties now deny this was finalized.
- The contract is part of a broader effort to develop AI tools for military use, including logistics, intelligence, and warfighting.
- Experts raise concerns about the role of private corporations in making determinations in warfare.
What happened
The Intercept obtained documents through a FOIA lawsuit revealing that OpenAI's Pentagon contract initially included a 'minimal refusal rates' clause for AI.
The contract, version 'P00003', is an update to a previous deal worth up to $200 million over two years.
Both OpenAI and the Pentagon have since denied that the 'minimal refusal rates' language was included in the final agreement.
The Pentagon initially confirmed the document's authenticity but later backtracked, stating it was a draft and not the final version.
Why it matters
The 'minimal refusal rates' clause raises concerns about the safeguards and ethical considerations in AI tools developed for military use.
The contract is part of a broader effort by the Pentagon to deploy AI tools across classified networks, involving companies like Google, xAI, and Anthropic.
Experts worry about the role of private corporations in making determinations in warfare, as seen in the public brawl between the Pentagon and Anthropic.
OpenAI claims it secured red lines on autonomous killings and spying against Americans, but the deal's drafting allows for any uses the government deems legal.
Key facts
- OpenAI's Pentagon contract initially included a 'minimal refusal rates' clause, but both parties deny this was finalized.
- The contract is worth up to $200 million over two years.
- The contract aims to develop AI tools for military use, including logistics, intelligence, and warfighting.
- The Pentagon initially confirmed the document's authenticity but later backtracked, stating it was a draft.
- OpenAI claims it secured red lines on autonomous killings and spying against Americans.
- The contract update includes a redacted section titled 'Testing, Evaluation, and Refinement of OpenAI Mission Models'.
- Experts raise concerns about the role of private corporations in making determinations in warfare.
- The Trump administration designated Anthropic a supply-chain risk, but the designation was overturned late last month by a federal judge.
Context
The Pentagon has been expanding its use of AI tools for military applications, involving companies like OpenAI, Google, xAI, and Anthropic.
The use of AI in warfare raises ethical considerations and concerns about the safeguards in place to prevent misuse.
The public brawl between the Pentagon and Anthropic highlights the tensions surrounding the use of AI in military applications and the role of private corporations in warfare.