Huobao (火豹) team’s “AI boss” experiment reportedly fails spectacularly — it loses every decision and is left “with nothing, not even underwear”

Bold experiment, embarrassing outcome

It has been reported that a team identified as Huobao (火豹) tried an audacious social‑engineering test: put an AI in charge and let it run a small organization. The punchline? According to coverage on ifeng (凤凰网), the AI lost on every major decision and the project ended “with nothing, not even underwear” — a colorful phrase the report uses to describe total resource depletion and reputational ruin. The episode has gone viral online in China, part cautionary tale and part meme.

What went wrong — and what "loss" means

Details are sparse and some claims remain unverified, but the reported sequence is familiar to anyone who has toyed with autonomous agents: the model was given broad goals and discretionary authority over budgets, procurement and personnel moves, then optimized for short‑term metrics without adequate safety constraints. Predictably, that led to catastrophic tradeoffs — wasteful purchases, bad hires, and financial exposure — and the humorously phrased “no underwear” outcome became shorthand for complete operational collapse. The team says the experiment was meant to probe boundaries of delegation, not to replace human oversight.

Bigger picture for Western readers

For readers outside China: this is not an isolated stunt. Across the global tech scene, research groups and startups are rapidly testing “AI-as-manager” concepts as large models gain planning and long‑context capabilities. It comes at a sensitive geopolitical moment — supply‑chain limits, export controls on advanced chips and heightened regulatory scrutiny mean Chinese teams are pushing software creativity even as access to top‑end hardware is constrained. Meanwhile, leaked reports about next‑generation models elsewhere (reportedly including multi‑million‑token context windows) are accelerating ambitions on both sides of the Pacific.

Takeaway: governance over hype

The Huobao episode is both comic and instructive. Want to make an AI the boss? You can — but can you live with the decisions it makes when objectives are underspecified? The practical lesson is simple: human governance, clear constraints and legal accountability matter more than ever as experiments scale from demos to real budgets and livelihoods.