Pentagon AI Warfare Timeline: Palantir, OpenAI & Anthropic Deals

As defense-technology equities open amid ongoing global tensions, the Pentagon’s AI integration has delivered measurable operational gains, including Palantir’s Maven Smart System now supporting near-real-time targeting across multiple combatant commands. The Department of Defense committed hundreds of millions in prototype contracts by mid-2025 alone, shifting from narrow experiments to classified frontier-model deployments. This timeline traces the sequence of decisions, contracts, and pivots that redefined AI’s role in warfare.

Table of Contents

• 2017–2018: Project Maven Launches AI Targeting
• 2022: CDAO Centralizes AI Efforts
• 2024: Palantir Secures Maven and TITAN Contracts
• 2025: Frontier Model Contracts with OpenAI and Anthropic
• January 2026: New AI Strategy and Pace-Setting Projects
• February 2026: Anthropic Dispute and OpenAI Pivot

2017–2018: Project Maven Launches AI Targeting

The Pentagon faced overwhelming video and sensor data from counter-ISIS operations. In April 2017, Deputy Secretary of Defense Bob Work signed the memo establishing the Algorithmic Warfare Cross-Functional Team, known as Project Maven. The initiative tasked commercial computer vision with identifying objects in full-motion video to accelerate targeting cycles.

• Google provided initial TensorFlow models under a $9 million pilot contract.
• Employee protests in 2018 prompted Google’s withdrawal from Maven follow-on work.
• The DoD assumed full control, proving AI could reduce analyst workload by orders of magnitude and establishing the template for future commercial partnerships.

This early success demonstrated causality: data overload forced rapid experimentation, but ethical pushback created the precedent for government-led development and stricter vendor oversight in later programs.

Key Contract Snapshot

2022: CDAO Centralizes AI Efforts

Fragmented AI initiatives across services created duplication and slow adoption. In December 2021 the Deputy Secretary announced the Chief Digital and Artificial Intelligence Office. By 2022 the CDAO reached full operational capability, absorbing Project Maven, the Joint AI Center, Advana data platform, and other efforts under one roof.

• Centralization enabled unified data standards and faster authorization-to-operate processes.
• The office began scaling Maven algorithms beyond imagery to multi-domain fusion.
• Consequence: DoD gained a single accountable entity for AI policy, paving the way for larger commercial contracts in subsequent years.

2024: Palantir Secures Maven and TITAN Contracts

Operational needs in the Indo-Pacific and Middle East demanded deeper sensor fusion. In March 2024 the Army selected Palantir for the Tactical Intelligence Targeting Access Node (TITAN) prototype phase at $178.4 million. Two months later, on May 29, 2024, the Army awarded Palantir a $480 million firm-fixed-price contract for the Maven Smart System prototype.

• Maven Smart System fuses ISR feeds, applies machine-learning for positive identification, and feeds fire missions directly into artillery systems.
• The platform already operated in U.S. Central Command for real-world targeting validation.
• Palantir’s role expanded Maven from experimental to production-grade, with contract ceiling later increased toward $1.3 billion by 2029, reflecting sustained demand across services.

These awards showed commercial data-analytics firms could deliver classified, battle-tested AI faster than traditional primes, setting the stage for frontier large-language-model integration.

2025: Frontier Model Contracts with OpenAI and Anthropic

Generative AI maturity prompted the CDAO to seek advanced reasoning capabilities. In June 2025 OpenAI received a contract worth up to $200 million to develop prototype frontier AI for both back-office and warfighting use. One month later, in July 2025, Anthropic secured a parallel two-year, $200 million prototype contract for “Claude Gov” variants optimized for classified environments.

• Contracts focused on intelligence synthesis, logistics forecasting, adversarial modeling, and simulation.
• Anthropic’s Claude became the first frontier model cleared for classified networks, integrated via Palantir infrastructure.
• OpenAI launched “OpenAI For Government” branding, signaling long-term defense alignment while both firms maintained internal safety policies.

The dual awards accelerated AI adoption but introduced governance friction that surfaced in 2026.

January 2026: New AI Strategy and Pace-Setting Projects

The new administration prioritized speed. On January 9, 2026 the Secretary of War released the Artificial Intelligence Strategy for the Department of War. The memorandum directed an “AI-first” force, re-focused the CDAO under the Under Secretary for Research and Engineering, and launched seven Pace-Setting Projects (PSPs).

• Warfighting PSPs included Swarm Forge, Agent Network, and Ender’s Foundry for simulations.
• Intelligence projects covered Open Arsenal and Project Grant for dynamic deterrence.
• Enterprise initiatives featured GenAI.mil and Enterprise Agents to democratize experimentation.

Timelines required initial demonstrations within six months and monthly progress reporting. Commercial model parity became policy: latest frontier releases must deploy within 30 days. The strategy explicitly leveraged private-sector innovation, directly benefiting contract holders like Palantir, OpenAI, and initially Anthropic.

February 2026: Anthropic Dispute and OpenAI Pivot

Negotiations over model usage rules reached impasse. On February 24, 2026 the Pentagon demanded unrestricted lawful use of Claude without internal guardrails. Anthropic refused to allow mass domestic surveillance or fully autonomous lethal weapons, citing constitutional and ethical constraints baked into model architecture.

• February 26: CEO Dario Amodei publicly rejected the “best and final” offer.
• February 27: The administration designated Anthropic a supply-chain risk and ordered phase-out within six months.
• February 28: OpenAI announced its classified-network agreement with explicit red lines matching Anthropic’s core prohibitions, plus cloud-only deployment and human-in-the-loop requirements for autonomous systems.

Palantir began unwinding Claude dependencies from Maven workflows. OpenAI assumed the replacement role within days, maintaining similar safeguards while enabling continued intelligence and simulation support. The episode underscored that commercial AI providers must balance Pentagon requirements with enforceable safety commitments.

If the Department of War sustains its “wartime CDAO” tempo and enforces uniform “any lawful use” contract language across vendors, commercial frontier-model companies aligned on classified integration will likely secure multi-year revenue streams through at least 2030.