University of Toronto Researchers Demonstrate AI‑Powered Worm That Autonomously Exploits Known Flaws

Key Points

• University of Toronto team built a prototype worm powered by open‑source AI models.
• Worm autonomously identifies and exploits known software vulnerabilities across Linux, Windows, and IoT devices.
• It harvests passwords and uses infected machines' processing power to improve its own attacks.
• AI enables the worm to switch to alternate flaws when a patch is applied, keeping infections alive.
• Researchers conducted experiments in a sealed, offline network to avoid real‑world impact.
• The work highlights a new class of AI‑driven cyber threats that could lower attack costs dramatically.
• Anthropic's Mythos AI has already uncovered over 10,000 vulnerabilities, showing the power of AI in security research.
• Security firms like Cloudflare are using AI tools to find thousands of flaws, including critical ones.
• Authors urge policymakers and industry to develop safeguards against AI‑enhanced malware.