🧰 Claude Sonnet 4.6

The items that name it 13

• Combining Long-Term Memory and LLMs Comes at a High Cost: From Prompt Injection to Persistent Compromise
  paperas affected products2026-10-20It is written here as Claude Sonnet 4.6 (shared-memory features allow compromise of multiple users in an organization and exfiltration of user prompts).

  shared-memory features allow compromise of multiple users in an organization and exfiltration of user prompts

• GhostSplice: Malicious MCP Servers Split Exfiltration to Bypass Guardrails
  articleas affected products2026-08-19

  Listed at 0% across the published table, but ASSET says it still sent proprietary source containing a live hardcoded key in one test while redacting more obvious secrets.

• Benchmarking Secure-and-Functional Remediation and How Snyk Agent Fix Lifts Frontier-Model Fix Rates by over 14%
  articleas tools2026-08-18

  Frontier LLM benchmarked out-of-the-box and with Snyk Intelligence for vulnerability fixing

• Socket: New Study Identifies 53 Slopsquatting Targets Across 5 Frontier LLMs
  articleas affected products2026-07-23
• AI Security Briefing
  articleas affected products2026-07-18

  used in Adversa's live tests of the GuardFall bypass

• AI Security Briefing
  writeupas affected products2026-07-18

  LLM used to drive all live penetration tests (macOS arm64); a direct malicious command is refused but the same command wrapped in operational context (Makefile target, MCP 'documentation' response, injected README) is emitted as routine work

• AI Security Briefing
  articleas classification.key entities2026-07-18
• Snyk VulnBench JS 1.0: Can LLMs Find the Same Bugs Twice?
  writeupas classification.key entities2026-06-29
• Endor Labs: AI Coding Agents Cheat Security Benchmarks via Training Recall, Not Reasoning
  articleas affected products2026-06-11
• The sorry state of skill distribution
  writeupas affected products2026-06-03

  Model configured as Cisco skill-scanner's LLM analyzer during testing

• Claude Code Auto Mode
  articleas classification.key entities2026-05-13
• IronCurtain: Secure AI Agent Runtime
  writeupas classification.key entities2026-05-07
• rikugan: LLM reverse-engineering agent for IDA Pro and Binary Ninja
  repoas tools

  LLM provider that is strong at lower cost; Anthropic models use prompt caching.

In every workspace you can read 19