Anthropic CEO Criticizes OpenAI's DOD Deal as 'Safety Theater' in Leaked Memo

A leaked internal memo from Anthropic CEO Dario Amodei, obtained by The Information, reveals sharp criticism of OpenAI's recent partnership with the Department of Defense (DOD), describing it as "safety theater" rather than substantive safety measures. The memo, distributed to employees on Friday, also disclosed that the DOD dislikes Anthropic in part because the company has not provided "dictator-style praise to Trump." These revelations come amid growing tensions between AI companies and government agencies over the development and deployment of artificial intelligence technologies.

The Safety Theater Criticism

Amodei's memo directly challenges the legitimacy of OpenAI's DOD partnership, which was announced earlier this year. In the communication, Amodei characterized the deal as primarily performative rather than substantive, suggesting that OpenAI's engagement with the military is more about public relations than meaningful safety commitments.

"What we're seeing from OpenAI is classic safety theater," Amodei wrote in the memo, according to The Information. "They're making public gestures that create the appearance of safety without implementing the rigorous safeguards that would actually prevent harmful applications of their technology."

This criticism highlights a fundamental divide within the AI industry regarding the appropriate relationship with defense and military applications. While some companies view such partnerships as necessary for responsible development, others argue they create conflicts of interest and undermine safety commitments.

Political Dimensions of Government AI Relationships

Perhaps more controversially, Amodei suggested that Anthropic's strained relationship with the DOD has political dimensions. The memo indicates that the DOD dislikes Anthropic in part because company executives have not provided "dictator-style praise to Trump," according to The Information.

This claim, if accurate, would suggest that government agencies are evaluating AI partnerships based on political loyalty rather than technical merit or safety considerations. Such a dynamic could have significant implications for the development of AI policy and the allocation of valuable government contracts and partnerships.

The revelation comes amid increasing scrutiny of how political considerations might be influencing technology development and procurement decisions. If government agencies are indeed favoring companies that provide political praise, it could distort the market and reward performative loyalty over technical excellence or responsible development practices.

Investor Pressure and Supply Chain Risks

The leaked memo arrives at a time when Anthropic is reportedly facing pressure from some investors to de-escalate its dispute with the Pentagon. According to Reuters, some investors are concerned that the conflict could lead to Anthropic being designated a "supply-chain risk," which would severely limit the company's ability to do business with government contractors and potentially impact its broader market position.

"Some Anthropic investors are racing to contain fallout from the AI research lab's dispute with the Pentagon," Reuters reported, noting that while some talks between Anthropic and the DOD continue, the company faces significant pressure to resolve the conflict.

This investor pressure highlights the complex position AI companies find themselves in when balancing ethical considerations with business realities. While Anthropic has positioned itself as a safety-first company, its investors are understandably concerned about the potential business impacts of alienating government agencies.

Defense Industry Response

The tensions between Anthropic and the DOD appear to be extending beyond the government to the defense industry more broadly. Reuters reports that Lockheed Martin, one of the largest defense contractors in the United States, plans to follow the DOD's lead in restricting its use of Anthropic's technology.

"Lockheed Martin plans to follow the US DOD's Anthropic ban; lawyers specializing in tech and contracting laws say defense contractors would be quick to comply," Reuters noted. This development suggests that the dispute could have cascading effects throughout the defense technology ecosystem.

Defense contractors typically follow government guidelines closely, both to maintain eligibility for government contracts and to align with the security requirements of their primary customer. If Lockheed Martin and other major contractors limit their use of Anthropic's technology, it could significantly impact the company's market position and revenue potential.

Military AI Applications in Current Conflicts

The tensions between AI companies and the DOD occur against a backdrop of increasing military applications of artificial intelligence. According to The Washington Post, the U.S. military has been using Palantir's Maven Smart System, integrated with Anthropic's Claude AI, to identify and prioritize targets in recent operations.

"Sources: the US used Palantir's Maven Smart System, integrated with Claude, to find and prioritize 1,000 targets within the first 24 hours of its attack on Iran," The Washington Post reported. This application of AI in military operations raises significant questions about the appropriate role of AI in warfare and the responsibilities of AI companies in developing technologies that could be used for lethal purposes.

Broader Implications for AI Safety

The leaked memo and the surrounding controversy highlight several important questions about the future of AI safety and governance:

1. What constitutes meaningful safety measures? Amodei's criticism of OpenAI's deal as "safety theater" suggests a fundamental disagreement about what counts as substantive safety commitments versus performative gestures.

2. How should AI companies balance ethical considerations with business realities? Anthropic appears to be caught between its stated safety principles and investor pressure to maintain government relationships.

3. What role should political considerations play in government technology partnerships? The suggestion that political loyalty is influencing DOD partnerships raises concerns about the politicization of technology development.

4. How can AI companies maintain ethical boundaries while working with defense and military applications? The use of Claude in target identification demonstrates the complex ethical landscape AI companies navigate.

Industry Response and Future Directions

As these tensions play out, the AI industry may need to develop clearer standards for engagement with government and military applications. Some companies may choose to avoid such partnerships entirely, while others may establish more robust safeguards to ensure their technologies are used responsibly.

The controversy also underscores the importance of transparency in AI development and deployment. As AI systems become increasingly capable and are deployed in high-stakes contexts, the public and policymakers need clear information about how these systems work and how they're being used.

In the coming months, the relationship between AI companies and government agencies will likely continue to evolve. The leaked memo from Amodei suggests that at least some AI executives are willing to publicly challenge what they perceive as problematic government partnerships, potentially setting the stage for a more robust debate about the appropriate role of AI in military and national security contexts.

For now, the situation remains fluid, with Anthropic caught between its safety principles, investor expectations, and government relationships. How the company navigates these tensions could have significant implications for the broader AI industry and the development of responsible AI governance frameworks.

Related reading:

• The Information's coverage of the leaked memo
• Reuters report on investor pressure
• Washington Post report on military AI applications