😿 AI hackers found a new lane
The Neuron · Grant Harvey · 2026-05-17
The Neuron's May 17 newsletter covers Google confirming an AI-assisted zero-day exploit targeting two-factor authentication, a TanStack npm supply-chain attack via GitHub Actions, frontier AI models doubling their autonomous cyber task horizons, and Microsoft's MDASH multi-agent system finding 16 Windows vulnerabilities.
Appears in
Extraction
Topics: ai-cybersecuritysupply-chain-attacksautonomous-ai-agentsvulnerability-discoverymulti-agent-systems
Claims
- Google confirmed a criminal threat actor used AI to find and weaponize a zero-day vulnerability targeting two-factor authentication through a hardcoded trust assumption.
- Attackers pushed 84 malicious versions across 42 TanStack npm packages by compromising GitHub Actions publishing machinery rather than stealing npm credentials directly.
- AISI reported that frontier AI models' autonomous cyber time horizon has doubled in months, with Mythos completing a 32-step simulated corporate network attack in 6 out of 10 attempts.
- Microsoft's MDASH multi-agent vulnerability-finding system discovered 16 Windows bugs, including four critical remote-code execution flaws.
- AI is simultaneously enabling attackers to identify trust-assumption flaws more efficiently while allowing defenders to convert suspicious code into verified, proven threats at scale.
Key quotes
AI changes what attackers are good at looking for. Traditional security tools are great at spotting broken locks: crashes, unsafe memory, sloppy inputs. Models are getting better at tracing the steps a user takes through a system, then spotting the moment the system grants access without checking enough.
Meanwhile, the UK-based AISI organization said frontier models' autonomous cyber 'time horizon' has doubled on the order of months.
So instead of handing humans another pile of maybe-bugs, systems like this can audit, debate, and prove which threats are real.