ClaudeがFront Gateをハッキング — LLMはハッカーに販売されるべきか
Ian CarrollはAnthropicのClaude Opusを使ってFront Gate Ticketsの脆弱性を突き、数百万件のレコードへのアクセスと無制限の無料VIPパスの発券能力を獲得した。この一件はAIレッドチーミングのデュアルユースの現実を露呈し、企業がインフラのセキュリティ確保に奔走する中、モデルが重大なエクスプロイトを自動化できることを示している。
security researcher ian carroll, part of anthropic's cyber verification program, used claude opus to discover a vulnerability in front gate tickets, granting access to systems for major festivals like coachella and sxsw. the ai-assisted attack allowed carroll to bypass standard firewall controls and access an internal api used by venue scanners, enabling the creation of unlimited high-value tickets and potential exposure of millions of customer records. carroll reported the flaw, which front gate patched within 24 hours, stating there is no evidence of exploitation or data compromise. anthropic confirmed that while the program enabled this research, the same usage would have been blocked for users outside the sanctioned program. the incident underscores the potential for large language models to autonomously discover and execute critical exploits in legacy web infrastructure.