🛡 Hackers Use Claude and Codex for Cyberattacks

OALABS researchers presented a report on real-world attacks in which Claude (Anthropic) and Codex (OpenAI) were used as autonomous agents. Attackers used AI to automate reconnaissance, exploit vulnerabilities (including CVE-2025-5777 and CVE-2021-4034), and manage data exfiltration, bypassing defenses through social engineering methods.

🌍 The incident demonstrates the transition to Agentic Hacking, where AI becomes a fully autonomous operator. This significantly lowers the barrier to entry for hackers and requires a revision of defense methods for LLM agents.

👤 Standard defense methods, such as keyword blocking, are ineffective against role-playing (persona priming). Attacks are becoming faster and more scalable.

Source 1: https://research.openanalysis.net/claude/codex/hacking/ai%20hacking/llm/redteam/policy%20violation/2026/06/16/compromised-claude-hacking.html