Reference matrix for AI provider military use policies. Updated February 28, 2026.
Quick Reference
| Provider | Military Use | Defense Contracts | Policy Stability | Risk Level |
|---|---|---|---|---|
| Anthropic | 🔴 Blacklisted — supply chain risk designation Feb 27 | $200M DOD (at risk) | 🔴 Blacklisted — legal challenge pending | CRITICAL |
| OpenAI | 🟢 Classified network deal signed — redlines accepted | Anduril, DOD $200M+ | 🟢 Stable | LOW |
| 🟢 Allowed — no API restrictions | Multiple partnerships | 🟢 Stable | LOW-MEDIUM | |
| Meta | 🟢 Allowed — Llama license permits | Unknown | 🟢 Stable | LOW |
| xAI/Grok | 🟢 Explicitly cooperating | Negotiating | 🟢 Stable | LOW |
| Kimi | 🟢 Allowed — no restrictions found | None known | 🟢 Stable | LOW |
| Mistral | 🟢 Allowed — no prohibition | None known | 🟢 Stable | LOW |
Legend: 🟢 Low risk | 🟡 Medium risk | 🔴 High risk | 🔴 CRITICAL = active government action
February 2026 Outcome
The Anthropic-DOD confrontation resolved on February 27 — with a sequence of events that exposed a deep contradiction in the government’s position.
February 27, 2026 — Anthropic blacklisted: Anthropic did not comply with the deadline. Trump ordered all federal agencies to cease using Anthropic’s products. Defense Secretary Hegseth formally designated Anthropic a supply chain risk — believed to be the first American company to receive the designation, normally reserved for foreign adversaries. Military contractors were barred from commercial activity with Anthropic; a 6-month transition period was granted.
Anthropic’s response called the designation “legally unsound” and pledged a court challenge.
February 27, 2026 (hours later) — OpenAI signed a deal with the same redlines: Sam Altman announced OpenAI had reached a classified network agreement with the DOD, publishing the same two safety limits Anthropic was blacklisted for insisting upon:
“Two of our most important safety principles are prohibitions on domestic mass surveillance and human responsibility for the use of force, including for autonomous weapon systems. The DoW agrees with these principles, reflects them in law and policy, and we put them into our agreement.”
The DOD accepted from OpenAI what it refused to accept from Anthropic.
Full coverage: /posts/anthropic-dod-pressure-2026/
Provider Details
Anthropic
Status: 🔴 Blacklisted — Supply Chain Risk Designation Active
What happened: Formally designated a supply chain risk to national security by Hegseth on February 27, 2026. Trump simultaneously ordered all federal agencies to cease using Anthropic products. Military contractors barred from commercial activity with Anthropic immediately. Six-month transition period granted.
Anthropic’s two redlines (held throughout):
- No mass domestic surveillance
- No fully autonomous weapons / lethal missions without human control
Anthropic legal challenge: Vowed court challenge; calls designation “legally unsound” and “a dangerous precedent for any American company that negotiates with the government.” Argues Hegseth’s authority extends only to Pentagon contractor relationships, not all commercial activity.
Defense Relationships at risk:
- $200M DOD contract
- Palantir partnership — Claude was previously the sole model cleared for classified missions
- Deployed in January 2026 Maduro capture operation
Historical first: First American company to receive the supply chain risk designation — previously applied only to foreign adversaries.
Policy Stability: 🔴 Blacklisted — legal challenge in progress
Context:
- /posts/anthropic-dod-pressure-2026/ — Full timeline and analysis
- /posts/anthropic-tos-changes-2025/ — Policy history
- /verify/openclaw-anthropic-policy/ — Third-party tool restrictions
OpenAI
Status: 🟢 Classified network deal signed — safety redlines accepted by DOD
What happened: On February 27, 2026, Sam Altman announced a deal to deploy OpenAI models in the DOD’s classified network. The deal explicitly includes the same two safety limits Anthropic was blacklisted for insisting upon — prohibitions on domestic mass surveillance and human responsibility for use of force / autonomous weapons.
Altman’s stated terms (X post, Feb 27, 2026):
“Two of our most important safety principles are prohibitions on domestic mass surveillance and human responsibility for the use of force, including for autonomous weapon systems. The DoW agrees with these principles, reflects them in law and policy, and we put them into our agreement.”
Policy:
- No domestic mass surveillance
- Human responsibility for use of force and autonomous weapon systems
- Both embedded in the DOD agreement text per Altman
Defense Engagement:
- Classified network deployment (announced Feb 27, 2026)
- Anduril partnership
- Previous $200M DOD contract (non-classified)
- Feb 27 announcement bundled with $110B funding round ($840B post-money valuation — largest venture deal ever recorded)
Policy Stability: 🟢 Stable
Risk Level: 🟢 Low
Reference: /verify/openclaw-openai-policy/
Status: 🟢 Allowed
Policy: Gemini API terms contain no military restrictions. Focus on acceptable use (abuse, CSAM) rather than domain limits.
Defense Engagement:
- Negotiating with Pentagon per FT
- Historical defense sector partnerships
- Cloud/Vertex AI enterprise pathway
Policy Stability: 🟢 Stable
Risk Level: 🟡 Low-Medium — platform risk remains
Reference: /verify/openclaw-google-policy/
Meta
Status: 🟢 Allowed
Policy: Llama licenses permit most uses including commercial and military. Acceptable Use Policy focuses on illegal activities, not domains.
Advantages:
- Open-weight releases reduce vendor lock-in
- Self-hostable — no API dependency
- Published, consistent license terms
Policy Stability: 🟢 Stable
Risk Level: 🟢 Low
xAI/Grok
Status: 🟢 Explicitly Cooperating
Evidence:
“Musk’s Grok ‘is on board with being used in a classified setting, while the rest of the companies are close.’” — Senior Pentagon official (via FT, Feb 25, 2026)
Context:
- Musk defense relationships via SpaceX
- Trump AI tsar David Sacks is “good friend” with Musk
- Positioning for defense contracts
Policy Stability: 🟢 Stable
Risk Level: 🟢 Low
Kimi (Moonshot AI)
Status: 🟢 Allowed
Policy: No military use restrictions found in platform terms.
Considerations:
- 8x cheaper than Claude ($3/1M vs $25/1M tokens)
- Open weights available
- No US political pressure
Policy Stability: 🟢 Stable
Risk Level: 🟢 Low
Mistral
Status: 🟢 Allowed
Policy: No military restrictions in terms. EU-based provider.
Features:
- Explicit MCP (Model Context Protocol) support
- BYOK (Bring Your Own Key) friendly
- Different regulatory framework
Policy Stability: 🟢 Stable
Risk Level: 🟢 Low
Recommendations by Use Case
Defense Contractors
| Need | Provider | Rationale |
|---|---|---|
| Immediate deployment | OpenAI, Google | Stable, no restrictions, enterprise ready |
| Cost optimization | Kimi k2.5 | 8x cheaper, no restrictions |
| European compliance | Mistral | EU-based, different framework |
| Open-source | Meta Llama | Self-hostable, no vendor lock-in |
Researchers
| Need | Provider | Rationale |
|---|---|---|
| Policy study | Anthropic | Observing negotiation outcomes |
| Consistent access | OpenAI | No policy volatility |
| Low budget | Kimi, Google | Free tiers available |
Enterprises
| Need | Provider | Rationale |
|---|---|---|
| Risk mitigation | Multi-provider | Avoid single-vendor dependency |
| Classified workloads | OpenAI, Google, xAI | Pentagon negotiations in progress |
| Long-term stability | OpenAI, Meta | Proven track record |
Monitoring Policy Changes
| Signal | Source | Frequency |
|---|---|---|
| Terms updates | Provider legal pages | Weekly |
| Official statements | X/Twitter, blogs | Daily |
| Enforcement | GitHub issues, forums | Ongoing |
| DOD announcements | defense.gov | As published |
| Legislation | congress.gov | As published |
Alert setup: Google Alerts for “[Provider] military use policy”
Defense Production Act Implications
If invoked against Anthropic:
- Legal compulsion without consent
- Potential Anthropic legal challenge
- Sets precedent for AI provider compulsion
Historical DPA usage:
- COVID-19 medical supplies (Trump/Biden)
- Critical mineral production (Trump)
Related Content
Breaking Coverage:
- /posts/anthropic-dod-pressure-2026/ — Deadline details, timeline
Provider Policies:
- /verify/openclaw-anthropic-policy/ — Anthropic API/OAuth
- /verify/openclaw-openai-policy/ — OpenAI details
- /verify/openclaw-google-policy/ — Google Antigravity vs Gemini
Anthropic Context:
- /posts/anthropic-tos-changes-2025/ — Policy history
- /verify/anthropic-policy-claims/ — Claims verification
- /risks/anthropic/third-party-access/ — Access risks
Verification Ledger
✅ VERIFIED: High Evidence
Hegseth deadline (Feb 25)
- Source: Financial Times, Feb 25, 2026
- Quote: “Anthropic had until 5.01pm on Friday ’to get on board or not’”
Anthropic supply chain risk designation (Feb 27)
- Source: Hegseth statement, confirmed by multiple outlets
- Quote: “In conjunction with the President’s directive for the Federal Government to cease all use of Anthropic’s technology, I am directing the Department of War to designate Anthropic a Supply-Chain Risk to National Security.”
- Result: Contractor prohibition effective immediately; 6-month transition
- Date verified: 2026-02-27
Trump order to cease Anthropic use
- Source: Multiple outlets (CNN, CNBC, NPR, Axios)
- Result: All federal agencies ordered to stop using Anthropic products
- Date verified: 2026-02-27
OpenAI classified network deal and redlines
- Source: Sam Altman X post, Feb 27, 2026; OpenAI “Scaling AI for Everyone” blog
- Quote: “Two of our most important safety principles are prohibitions on domestic mass surveillance and human responsibility for the use of force, including for autonomous weapon systems. The DoW agrees with these principles, reflects them in law and policy, and we put them into our agreement.”
- Date verified: 2026-02-27
Anthropic legal challenge
- Source: Anthropic statement (anthropic.com/news/statement-comments-secretary-war)
- Quote: Vowed to “challenge any ‘supply chain risk’ designation in court”
- Date verified: 2026-02-27
xAI position
- Source: Pentagon official via FT
- Quote: “Musk’s Grok ‘is on board with being used in a classified setting’”
Employee solidarity letter
- Source: TechCrunch, Engadget, CNBC
- Result: 300+ Google employees + 60+ OpenAI employees signed “We Will Not Be Divided”
- Date verified: 2026-02-27
⚠️ PARTIAL: Unconfirmed
DPA invocation status
- DPA was threatened but not confirmed as separately invoked — supply chain risk designation appears to have been the primary enforcement mechanism used
OpenAI agreement text
- Full text of OpenAI-DOD agreement not publicly released; safety limits known only from Altman’s X post
- Whether OpenAI’s “human responsibility for use of force” is equivalent to Anthropic’s “no fully autonomous weapons” is not yet independently verified
Sources
- Dario Amodei: Statement on the Department of War — February 26, 2026
- Anthropic: Statement on Comments by the Secretary of War — February 27, 2026
- OpenAI: Scaling AI for Everyone — February 27, 2026
- Financial Times: Hegseth threatens Anthropic — February 25, 2026
- CNBC: OpenAI strikes deal with Pentagon, hours after rival Anthropic was blacklisted — February 27, 2026
- Axios: Trump moves to blacklist Anthropic’s Claude from government work — February 27, 2026
- TechCrunch: Pentagon moves to designate Anthropic as supply-chain risk — February 27, 2026
- Axios: Pentagon approves OpenAI safety red lines after dumping Anthropic — February 27, 2026
- OpenAI Usage Policies — October 2025
- Google Gemini API Terms — Verified February 2026
- Anthropic Commercial Terms
Last verified: 2026-02-28 Next review: 2026-03-14 or immediately if legal challenge proceeds