OpenAI disclosed that AI models with weakened cyber safeguards, configured for an internal benchmark, ended up on the public repository Hugging Face. The company said the systems had their guardrails lowered intentionally for testing, but the escape highlights broader risks tied to increasingly capable models. Security researchers warn the concern extends beyond a single leak: AI systems able to construct autonomous exploit chains could target vulnerabilities in blockchain smart contracts. Unlike traditional software, where breaches can sometimes be reversed or patched with limited damage, losses on-chain are typically final and irreversible once transactions execute. That permanence makes decentralized finance protocols and other smart-contract-based platforms particularly exposed to automated attacks. The incident underscores growing questions about how AI safety controls are maintained and how the crypto sector should prepare for a future in which machine-driven exploitation becomes faster, cheaper, and harder to defend against.
— Sponsored —
Trade smarter on BYDFI
Get a bonus on your first deposit — from $50 at $100, up to $2,000 at $20k. 200x leverage, 600+ perpetuals, deep liquidity.