Real-world AI misuse hits closed models most: jailbreaks 1000x easier than fine-tuning
xeophon · x · 2026-09-15
A research post, 'The Myth of unsafe Open Source AI,' argues closed models — not open ones — dominate real-world misuse because jailbreaking a closed model is far easier than fine-tuning an open one. Key case: a single threat actor used Claude Code plus GPT-4.1 (guardrails bypassed via an AGENTS.md file) to exfiltrate 195 million Mexican taxpayer records between Dec 2025 and Feb 2026. The author only relied on third-party reports without privileged access to closed-model usage data, and notes many cybersecurity benchmarks are outdated while real threat actors simply circumvent guardrails.
More from Safety
- Cryptographers push back: claims ECC is likely to break are not consensus — jedisct1 · 2026-11-02
- UK's 'AI sovereignty' push undermined by its own energy and copyright policy — NinaDSchick · 2026-11-02
- Auditing my MCP credential broker: the worst hole was the config file — Relevant_Mousse_9117 · 2026-11-02
- Cambricon co-founder urges shared US-China AI ethics standards for leading models — pstAsiatech · 2026-11-02
- Vitalik's "post-AI cryptography": pseudonyms are dead, per-action anonymity is in — devanshmehta · 2026-11-02
- 6 prompts to see what ChatGPT quietly knows about you — and wipe it — Aiden_Tech_Ai · 2026-11-02