OpenAI was actually lagging behind the Pentagon in using artificial intelligence for surveillance and military activities

OpenAI and the Pentagon: How the AI Deal Has Changed

*OpenAI has entered a new agreement with the U.S. Department of Defense (Pentagon), replacing Anthropic, but in response to public concerns its CEO Sam Altman stated that the terms were “set” in accordance with safety principles. In practice the situation looks different, as reported by The Verge.*

1. What Altman Promised
- Two key requirements:

1) a ban on mass internal surveillance;
2) human accountability for force use, including autonomous armed systems.

- The Pentagon agreed to “include” these points in law and policy, and OpenAI – in the contract.

2. Why Doubts Remain
- Social media raised questions: was there really an agreement to enforce the limits?
- Experts believe the Pentagon did not give a full rejection. The agreement contains the phrase “any lawful use,” which opens space for a wide range of actions.

3. What Was Actually Discussed
Side Action Key Point Pentagon Collect and analyze large volumes of data on U.S. citizens Did not back down from wanting to use data at scale OpenAI Restrictions on using the system for mass surveillance Stated that “you cannot collect or analyze American data in large, unlimited volumes.”
- The parties disagreed on how to interpret the term *technically lawful*.

- In the U.S., this concept is gradually expanding and now includes mass‑surveillance programs.
- OpenAI rejected that interpretation.

4. Legal Framework
- Main regulations:

- The Fourth Amendment;
- National Security Act (1947);
- Foreign Intelligence Surveillance Act (1978);
- Executive Order No. 12333;
- Pentagon’s internal directives on foreign intelligence.

- After 9/11 U.S. intelligence agencies intensified surveillance, claiming everything was within the law.
- Snowden (2013) revealed large‑scale phone‑call collection and data from Verizon, Microsoft, Google, Apple. EO 12333 allows interception of data outside the U.S., even about citizens.

5. Reporting Dialogue
Company Position Anthropic (Dario Amodei) Law does not yet cover all AI capabilities for mass surveillance. OpenAI (Sam Altman) Contract is limited by existing law; with new norms, wording will remain unchanged.
- Experts note that the words used (“unlimited,” “generic,” “open”) do not create a clear ban and may allow broad use of AI in intelligence.

6. Conclusions
- Parties: OpenAI claims compliance with safety principles, but the Pentagon retains the right to extensive surveillance capabilities.
- Law: The agreement relies on existing laws, but their interpretation could alter the reality of AI use in defense.
- Public discourse: Questions about transparency and accountability remain open.