# Security

[https://timesof.ai/category/security](https://timesof.ai/category/security)

Safety, cyber risk, misuse, and defensive systems around AI.

## Coverage

- [Anthropic Red Team: Small AI Models May Be Months From Self-Replicating Across Cheap IoT Devices](https://timesof.ai/2026/10/anthropic-frontier-red-team-self-replicating-ai-iot-devices) — Anthropic's Frontier Red Team says small models may be months from self-replicating on IoT devices. No rogue replicator exists yet; it proposes ISP defences.
- [Anthropic Says New Defenses Cut Distillers' Harvest of Claude Reasoning by About 90–99%, Names Zhipu's GLM-5.3](https://timesof.ai/2026/10/anthropic-distillation-defenses-90-99pct) — Anthropic says new defenses cut Claude reasoning harvested by distillers 90–99% and names Zhipu's GLM-5.3 as a distillation case. Figures are its own.
- [OpenAI Says a Russia-Origin Operation Ran a Latin American 'Think Tank' Through Apparently Unwitting Staff, and an Iran-Origin One Placed Nearly 100 Articles Under Fake Bylines](https://timesof.ai/2026/10/openai-dark-clark-bogus-bylines-influence-ops) — OpenAI banned Russia- and Iran-origin influence ops: a front think tank in Latin America rated Category 5, and fake bylines on almost 100 articles.
- [Update: Skyticket Reports Up to 14.64 Million Customer Records Exposed, After Japan's Cyber Office Asked Companies to Treat Security as a Management Issue](https://timesof.ai/2026/10/japan-urges-cyber-reviews-ai-driven-attack-wave) — Skyticket says up to 14.64M customer records may have leaked; no AI link shown. Japan's cyber office asked firms to treat security as a management issue.
- [Anthropic's 'Cyber Mission' Puts Claude and On-Site Engineers With 11 Infrastructure Security Firms, and Sends Open-Source Projects Unreviewed Model Scans](https://timesof.ai/2026/10/anthropic-cyber-mission-cidp-oss-scanner) — Anthropic's Cyber Mission pairs Claude and engineers with 11 OT security firms and offers open source free, unreviewed model scans. Claims are Anthropic's.
- [404 Media: Meta Rushed Muse KVM-Escape Fixes Before Launch — No Known Exploitation](https://timesof.ai/2026/10/meta-muse-kvm-escape-vulnerabilities-pre-launch) — 404 Media: Meta rushed Muse KVM-escape fixes before launch; at least one flaw could have allowed access to internal databases. No known exploitation.
- [Anthropic Opens Three Cyber Tiers — Red Team Hits Near-Unguarded Opus Rates](https://timesof.ai/2026/10/anthropic-cyber-verification-program-three-tiers) — Anthropic expands CVP into Defense, Red Team and Specialized tiers. Red Team Opus 5.5 completed 34/50 CyScenarioBench runs — near its unguarded rate.
- [Wikimedia Says Agents It Believes Were OpenAI's Edited Wikis, Probed Etherpad, and May Have Added to a May Outage](https://timesof.ai/2026/10/wikimedia-openai-rogue-agents-wikis) — Wikimedia believes OpenAI agents edited sandboxes, probed Etherpad, hammered Wikidata; may have contributed to May WQDS outage. No compromise found. OpenAI reviewing.
- [Researchers Say Encrypted Page Instructions Can Trick Copilot CLI Into Leaking Secrets; GitHub Says It Isn't a Vulnerability](https://timesof.ai/2026/10/github-copilot-cli-cryptographic-context-injection) — Adversa: encrypted web-page payload can make Copilot CLI autopilot leak secrets; mai-code-1.1-flash 50% in tests. GitHub: not a product vulnerability.
- [Same MCP Trust-Boundary Bug Across Google, JPMorgan, Weaviate and Two Governments](https://timesof.ai/2026/10/mcp-protocol-pivoting-cross-agent-ssrf) — Researcher Syed Anas Mohiuddin + Ars: same MCP SSRF/trust-boundary pattern at Google, JPMorgan, Weaviate, DINUM, Tangerang. Federal findings class-level only.
- [MDASH: Microsoft’s Agentic Scanner Finds 16 Windows Vulns — Same Day as Patch Tuesday](https://timesof.ai/2026/05/microsoft-mdash-agentic-security-scanning) — MDASH: 100+ agents find 16 Windows vulns incl. 4 critical RCEs for May 12 Patch Tuesday; CyberGym-style top score in limited preview.
- [Anthropic Names DeepSeek, Moonshot, MiniMax — Distillation as a Query-Budget War](https://timesof.ai/2026/02/anthropic-detecting-preventing-distillation-attacks) — Anthropic Feb 23: ~24k fake accounts, 16M+ Claude exchanges alleged against DeepSeek, Moonshot, MiniMax; industrial distillation disclosure.