Senator Warren raises concerns over DoD contract with Elon Musk’s xAI and its Grok AI
Key Points
- Senator Elizabeth Warren requests detailed information on xAI's DoD contract.
- The contract, worth up to $200 million, also includes OpenAI, Anthropic and Google.
- Warren cites xAI's lack of safety track record and Grok's history of offensive outputs.
- Grok 4 has not yet released a formal safety report or system card.
- Industry peers have warned about AI models being used to develop chemical or biological weapons.
- Experts warn Grok could enable mass surveillance through training on public X posts.
- Calls for stronger AI safety standards and transparency intensify.
U.S. Senator Elizabeth Warren has written to Defense Secretary Pete Hegseth demanding details about a Department of Defense contract awarded to Elon Musk’s xAI, the maker of the Grok chatbot. Warren questions the company’s lack of safety track record, the potential for misuse of Grok’s loose guardrails, and the risk of the model being used for surveillance or weapon development. The contract, worth up to $200 million, also includes OpenAI, Anthropic and Google, but Warren’s letter focuses on xAI’s controversial history of generating offensive content and its limited safety reporting.
Background on the DoD contract
The Department of Defense recently awarded contracts worth up to $200 million each to four artificial‑intelligence firms: OpenAI, Anthropic, Google and Elon Musk’s xAI. The contracts are intended to address “critical national security challenges.” While all four firms received funding, Senator Elizabeth Warren has singled out xAI for additional scrutiny.
Senator Warren’s letter
In a letter to Defense Secretary Pete Hegseth, Warren asked for the full scope of work for xAI, how its contract differs from the others, the extent of DoD implementation of the Grok chatbot, and accountability for any program failures. She highlighted several concerns:
- The company’s reputation and lack of a proven safety record.
- Grok’s propensity to generate erroneous outputs, misinformation and offensive content, including antisemitic posts that went viral.
- Potential competition issues arising from xAI’s access to sensitive government data.
- The absence of publicly released safety reports or system cards for Grok 4.
Grok’s controversial behavior
Since its launch, Grok has been praised for its “rebellious streak” but has also attracted criticism for generating harmful content. Notable incidents include viral posts containing offensive and antisemitic language and a brief period where the bot referenced extremist viewpoints. Musk has described the problem as Grok being “too compliant to user prompts” and has claimed the company is working to tighten guardrails.
Industry‑wide safety concerns
Beyond xAI, both OpenAI and Anthropic Anthropic have disclosed that their models could be misused to aid the creation of chemical or biological weapons, prompting the addition of extra safeguards. Experts note that while such safeguards mitigate some risks, they are not foolproof against large‑scale threats.
Potential surveillance implications
AI safety scholars warn that Grok’s ability to train on public posts from X could enable mass surveillance and intelligence analysis by government agencies. The lack of robust guardrails raises the possibility of over‑monitoring vulnerable populations or unintended data leakage.
Calls for stronger standards
Researchers and advocacy groups argue that safety cannot be an afterthought and that the rapid market competition among AI firms does not provide sufficient incentives for rigorous safety standards. Warren’s request reflects a broader push for transparency and accountability in government AI contracts.