OpenAI is working with the US military now

Original Article ↗ Hacker News Discussion ↗

OpenAI’s new collaboration with the U.S. Department of Defense on open-source cybersecurity tools and veteran suicide prevention is seen as a sharp shift from its earlier pledge to avoid military work. Commenters are split between viewing this as a pragmatic, even necessary use of U.S. AI leadership and warning of a slippery slope toward weaponized, autonomous systems and an AI-driven arms race. The exchange broadens into questions about American hegemony, great-power competition, and whether any major AI company can realistically stay out of defense applications.

Scope of OpenAI–Pentagon Collaboration

  • Tools being developed for the US Defense Department focus on:
    • Open-source cybersecurity (via DARPA’s AI Cyber Challenge).
    • Support for preventing veteran suicide, likely in coordination with Veterans Affairs.
  • OpenAI reiterates a formal ban on using its tech for weapons, property destruction, or direct harm.
  • Some commenters note the article was updated to clarify this is DARPA/VA-related, and argue the headline is misleading.

Change from Original “No Military” Stance

  • Many see this as a sharp reversal from OpenAI’s earlier, explicit prohibition on military use.
  • Some argue this was always inevitable once money and power were at stake.
  • A few suggest the earlier stance may have been a recruiting/PR tactic to attract talent uncomfortable with defense work.

Slippery-Slope and AI Risk Concerns

  • Strong worry about incremental drift:
    • “Defensive tools” → “defensive weapons” → offensive systems.
  • Discussion of how adding autonomy, planning, and OODA-loop capabilities could make military AI an existential risk.
  • Others dismiss LLM chatbots themselves as non-autonomous and not an immediate existential threat.

Ethics of Defense Tech and US Hegemony

  • Split views:
    • Some welcome US military leadership in AI, seeing it as necessary in great-power competition and preferable to rivals.
    • Others fear an arms race, escalation risks, and long-term blowback; some explicitly desire an end to US hegemony.
  • Extended debate over whether the world is unipolar or multipolar (US vs EU/China/Russia/India), including arguments about:
    • Russia’s and China’s military and economic strength.
    • Proxy wars (Ukraine), sanctions, and war economies.
    • The US record in Afghanistan and limits of military power in nation-building.

Mental Health Chatbots for Veterans

  • Some see 24/7 chatbot access as a useful supplement to human care, with research suggesting benefit.
  • Others argue routing acutely suicidal people to a bot would feel dehumanizing and dangerous, especially given likely poor government implementation.

Capabilities, Misuse, and Prompt Injection

  • Skepticism about using current models for real cybersecurity/sysadmin tasks due to hallucinated commands.
  • Speculation about future battlefield prompt-injection tricks (e.g., messages on rooftops to confuse targeting models), framed as another iteration of cat-and-mouse in warfare.

Disillusionment with OpenAI’s Mission

  • Several commenters express that OpenAI has rapidly shifted from a mission-driven, capped-profit “don’t build weapons” organization to a conventional tech company aligned with state power.