Sources and Further Reading

Figures in this guide come from the sources below, checked in November 2026. Where only press coverage of a report was available, the coverage is listed. Several figures come from vendor-commissioned research and are labeled as such in the text. Check the originals before quoting figures in regulated or contractual contexts.

Anthropic, Mythos and Project Glasswing

• Anthropic: Project Glasswing (April 7, 2026): launch, partners, usage credits and open-source donations
• Anthropic red team: assessing Claude Mythos Preview's cybersecurity capabilities (April 7, 2026): OpenBSD, FFmpeg and Firefox findings, exploit chaining, method and cost
• Anthropic: Project Glasswing, an initial update (May 22, 2026): 10,000+ high- or critical-severity findings, 1,752 assessed findings
• Anthropic: expanded Cyber Verification Program (October 6, 2026): access tiers, 129,000+ verified vulnerabilities, CyScenarioBench results
• Anthropic red team: universal ECDSA certificate forgery in wolfSSL (CVE-2026-5194)
• VentureBeat: 6,157 findings reported to maintainers and 516 patched
• Decrypt: Claude Mythos leaks (March 2026)
• TechCrunch: Anthropic scales Claude Mythos to critical infrastructure in 15+ countries (June 2, 2026)
• MacRumors: Anthropic launches Claude Fable 5 (June 9, 2026)
• VentureBeat: Anthropic brings back Claude Fable 5 after the U.S. lifts its export control order
• eWeek: Claude Fable 5.1 and Mythos 5.1 (September 2026)
• AISLE: AI cybersecurity after Mythos, the jagged frontier
• Flying Penguin: critique of Glasswing verification (June 2026), a counterpoint on reproducibility

Threat and breach data

• CrowdStrike: 2026 Global Threat Report press release (February 24, 2026)
• Google Cloud: M-Trends 2026
• Verizon: 2026 Data Breach Investigations Report, with Help Net Security's summary
• IBM: Cost of a Data Breach Report (2026 edition), with Baker Donelson's ten takeaways
• The Register: VulnCheck on exploitation of AI-found vulnerabilities
• VulnCheck: tracking Anthropic and Glasswing CVEs
• Veracode: 2026 GenAI Code Security Report
• CISA: Known Exploited Vulnerabilities catalog
• FIRST: Exploit Prediction Scoring System (EPSS)
• Google: OSS-Fuzz

Hugging Face incident

• Help Net Security: Hugging Face breached by autonomous AI agent (July 20, 2026)
• Cloud Security Alliance: research note on the Hugging Face autonomous agent breach
• VentureBeat: safety guardrails blocked Hugging Face's defenders, not the attacker
• BleepingComputer: Hugging Face breach coverage

Machine identity and AI agents

• Palo Alto Networks: 2026 Identity Security Landscape report, with Help Net Security's summary
• Cloud Security Alliance: survey finds 82% of enterprises have unknown AI agents (April 21, 2026; commissioned by Token Security)
• Teleport: 2026 State of AI in Enterprise Infrastructure Security (February 2026)
• arXiv: Agents of Chaos, red-teaming autonomous LLM agents (February 2026)
• OWASP GenAI Security Project: Top 10 for Agentic Applications
• NIST: AI Agent Standards Initiative

Related reading on guptadeepak.com

• Credential lifecycle management for AI agents
• MCP, RAG and ACP: a comparative analysis
• Implementing FIDO2 authentication
• Password hashing: Argon2, bcrypt, scrypt and PBKDF2 compared
• The browser security landscape in 2025
• Agent Security Map, the identity-first landscape of AI agent security vendors
• The CISO Desk
• Hash Lab

About the author

Deepak Gupta founded LoginRadius in 2013 and scaled it to serve more than 1 billion users, helping define what is now called customer identity and access management (CIAM). In 2024 he started GrackerAI, which does Generative Engine Optimization for B2B cybersecurity vendors. He writes about AI, cybersecurity and identity at guptadeepak.com.