Skip to content
Submit incident
AII-1000-80031
ReportedReported by fewer than three independent publishers, with no official finding.

Researchers use Anthropic's Mythos to create macOS M5 security exploit

Cybersecurity firm Calif used Anthropic's Claude Mythos Preview AI model to develop a working exploit against Apple's M5 chip Memory Integrity Enforcement protections in less than a week. The exploit links two macOS bugs to achieve privilege escalation, demonstrating that AI can assist in discovering software vulnerabilities, though human expertise remained necessary.

1 May 2026 Β· Calif Β· Security and cyberattacks
1 May 2026
AII-1000-800312 sources ↓
Quick facts
Category
Security and cyberattacks
Country
United States
Source
MacRumors
Date
1 May 2026
Drafted by
Raidu Incident Detection

What happened

Calif, a cybersecurity startup based in Palo Alto, used Anthropic's Claude Mythos Preview model to develop a working exploit against Apple's M5 chip Memory Integrity Enforcement (MIE) protections. The exploit links two macOS bugs using techniques that corrupt the Mac's memory and gain access to system parts that should remain inaccessible.

According to Calif, the exploit was developed in less than a week, with bugs found on April 25th and a working exploit completed by May 1st. Calif described the discovery as the first public macOS kernel memory corruption exploit capable of surviving Apple's MIE protections.

Reported impact

The exploit demonstrates that AI can assist in bypassing security protections that Apple spent nearly five years developing and likely invested billions of dollars into. The vulnerability poses a potential risk to users of Apple's M5 hardware.

Index analysis

The use of Claude Mythos Preview was part of Anthropic's Project Glasswing, an initiative enabling tech companies to use the frontier AI model to find security vulnerabilities. Researchers stated the exploit would not have been possible with Mythos alone, as it required their human expertise in combination with the AI assistance.

Governance controls

Apple said it was reviewing Calif's report to validate the findings and that security is its top priority. Apple's macOS 26.5 security notes mention a fix for a kernel-level vulnerability crediting Calif and Anthropic for discovering it.

Classification

Category
Security and cyberattacks
Industry
Digital security
Country
United States
Harm status
Potential harm
Intent
Unintentional
Timing
Unknown
Harm types
Public interest
Harmed
Business
Consequences
None recorded
AI systems
Claude Mythos Preview

Sources and evidence

This record was researched and written by the Index, based on the following original reporting.

MacRumors
Apple Alerted to macOS Security Vulnerability Uncovered With AI Tool
14 May 2026
indiatoday.in
Researchers use Anthropic's Mythos to create macOS M5 security exploit