Round 20 April 7–8, 2026
- Source ID:
src-20260409-f5e09e325670 - Kind:
analysis - Scope:
shared - Origin:
community-analysis - Raw path:
sources/raw/round-20-april-7-8-2026__src-20260409-f5e09e325670.md - Status:
active
Summary
Community analysis document covering Claude Code architecture, internals, and leak analysis. See extracted entities and claims below.
Tags
analysis
Extracted entities
ent-20260409-d3922fa59086Operon Desktop Mode (service): Unreleased science research workspace for biology/healthcare discovered March 27ent-20260419-a1b2c3d4e501CVE-2026-35020 (issue): Unpatched critical (CVSS 8.4) command injection in Claude Code's which.ts commanent-20260419-a1b2c3d4e502CVE-2026-35022 (issue): Unpatched credential exfiltration vulnerability in auth.ts credential-helper patent-20260419-a1b2c3d4e503CVE-2026-35021 (issue): High-severity (CVSS 7.8) command injection in promptEditor.ts editor-launch pathent-20260419-a1b2c3d4e504Phoenix Security Kill Chain (concept): Single exploitation chain linking CVE-2026-35020 (TERMINAL env injection, zero uent-20260419-a1b2c3d4e505CVE-2026-31861 (issue): CVSS 8.8 command injection in the @siteboon/claude-code-ui adjacent-tooling packent-20260419-a1b2c3d4e506CVE-2026-34714 (Vim tabpanel RCE) (issue): CVSS 9.2 remote code execution in Vim discovered by Claude Code in ~2 minutes frent-20260419-a1b2c3d4e507GNU Emacs Git Forever-Day (issue): Zero-day dating back to 2018 in how GNU Emacs interacts with Git; exploitable sient-20260419-a1b2c3d4e508Linux Kernel 23-Year Vulnerability (issue): Linux kernel vulnerability present for ~23 years, surfaced by Nicholas Carlini (ent-20260419-a1b2c3d4e50912 Boring Primitives Framework (concept): Nate Jones' architectural framing of Claude Code as '20% LLM, 80% plumbing,' orgent-20260419-a1b2c3d4e50aNate Jones (person): Analyst whose YouTube breakdown and Substack deep-dive of the Claude Code leak pent-20260419-a1b2c3d4e50bNicholas Carlini (person): Research scientist at Anthropic. Used Claude Code to surface a 23-year-old Linuxent-20260419-a1b2c3d4e50cHung Nguyen (Calif AI Red Teaming) (person): Security researcher at Calif AI Red Teaming who sent Claude Code a one-sentenceent-20260419-a1b2c3d4e50dDark Code Thesis (concept): Forbes-coined category for AI-agent-authored code that compiles, passes tests, aent-20260419-a1b2c3d4e50eCVE-2026-2796 Autonomous JS Engine Exploit (document): Published on red.anthropic.com/2026/exploit/. Claude autonomously wrote a full Went-20260419-a1b2c3d4e50fJonny Teardown (document): Engineer 'Jonny (good kind)' teardown of the leaked Claude Code source, elevatedent-20260419-a1b2c3d4e510Anthropic IPO Timeline (Q4 2026) (concept): Anthropic's reported Q4 2026 (October) IPO target at $380B valuation (post-$30Bent-20260419-a1b2c3d4e511Anthropic VDP 'Informative' Disposition (decision): Anthropic Vulnerability Disclosure Program disposition closing CVE-2026-35020 an
Extracted claims
clm-20260419-a1b2c3d4e601[fact]: Three CVEs on v2.1.91 chain into no-click credential+MEMORY.md exfiltrationclm-20260419-a1b2c3d4e602[pattern]: All five known Claude Code CVEs share CWE-78 (OS command injection via unsanitized string interpolation)clm-20260419-a1b2c3d4e603[metric]: Claude Code discovered a Vim RCE zero-day in 2 minutes from a one-sentence promptclm-20260419-a1b2c3d4e604[fact]: Claude Code surfaced a 23-year Linux kernel vulnerability from an open-ended promptclm-20260419-a1b2c3d4e605[hypothesis]: Claude Code is 20% LLM call and 80% plumbing (12 primitives)clm-20260419-a1b2c3d4e606[fact]: Permission tiers in Claude Code evaluate per tool call, not per sessionclm-20260419-a1b2c3d4e607[pattern]: Workflow state and conversation state are stored separately so compaction preserves task memoryclm-20260419-a1b2c3d4e608[constraint]: Token budget is checked before each turn, not afterclm-20260419-a1b2c3d4e609[pattern]: Tool calls emit events to all listeners simultaneously — no pollingclm-20260419-a1b2c3d4e60a[pattern]: Verification is a separate model call, not a self-check in the main loopclm-20260419-a1b2c3d4e60b[fact]: Claude Code does not load all 40 tools into every contextclm-20260419-a1b2c3d4e60c[fact]: Six role-specialized agent types each have their own context budget and tool subsetclm-20260419-a1b2c3d4e60d[observation]: Anthropic invested in what Claude Code can do, not in what it producesclm-20260419-a1b2c3d4e60e[observation]: Undercover Mode makes dark code invisible to AI-attribution audit toolingclm-20260419-a1b2c3d4e60f[fact]: Claude autonomously constructed addrof+fakeobj from WebAssembly GC type confusionclm-20260419-a1b2c3d4e610[metric]: Claude Code at $2.5B ARR drives 80% enterprise share within a $19B Anthropic ARRclm-20260419-a1b2c3d4e611[decision]: Anthropic declined to patch TERMINAL/apiKeyHelper chain, citing 'controlling TERMINAL implies code execution'clm-20260419-a1b2c3d4e612[metric]: 18 independent security modules protect the Bash tool alone
Extracted relations
rel-20260419-a1b2c3d4e701: CVE-2026-35020 --[contains]--> Phoenix Security Kill Chainrel-20260419-a1b2c3d4e702: CVE-2026-35022 --[contains]--> Phoenix Security Kill Chainrel-20260419-a1b2c3d4e703: Phoenix Security Kill Chain --[depends_on]--> CVE-2026-35020rel-20260419-a1b2c3d4e704: Anthropic VDP 'Informative' Disposition --[blocks]--> CVE-2026-35020rel-20260419-a1b2c3d4e705: Anthropic VDP 'Informative' Disposition --[blocks]--> CVE-2026-35022rel-20260419-a1b2c3d4e706: Phoenix Security --[owns]--> Phoenix Security Kill Chainrel-20260419-a1b2c3d4e707: Hung Nguyen (Calif AI Red Teaming) --[caused]--> CVE-2026-34714 (Vim tabpanel RCE)rel-20260419-a1b2c3d4e708: Claude Code --[caused]--> CVE-2026-34714 (Vim tabpanel RCE)rel-20260419-a1b2c3d4e709: Claude Code --[caused]--> GNU Emacs Git Forever-Dayrel-20260419-a1b2c3d4e70a: Nicholas Carlini --[caused]--> Linux Kernel 23-Year Vulnerabilityrel-20260419-a1b2c3d4e70b: Claude Code --[caused]--> Linux Kernel 23-Year Vulnerabilityrel-20260419-a1b2c3d4e70c: Nate Jones --[owns]--> 12 Boring Primitives Frameworkrel-20260419-a1b2c3d4e70d: 12 Boring Primitives Framework --[summarizes]--> Claude Coderel-20260419-a1b2c3d4e70e: 12 Boring Primitives Framework --[supports]--> 7-Agent Feature Modelrel-20260419-a1b2c3d4e70f: 12 Boring Primitives Framework --[supports]--> Bash Security Pipelinerel-20260419-a1b2c3d4e710: Dark Code Thesis --[depends_on]--> Undercover Moderel-20260419-a1b2c3d4e711: Jonny Teardown --[related_to]--> CVE-2026-35020rel-20260419-a1b2c3d4e712: Jonny Teardown --[related_to]--> CVE-2026-35022rel-20260419-a1b2c3d4e713: CVE-2026-2796 Autonomous JS Engine Exploit --[related_to]--> Anti-Distillation Defensesrel-20260419-a1b2c3d4e714: Anthropic IPO Timeline (Q4 2026) --[informed_by]--> Source Map Leak (March 31, 2026)rel-20260419-a1b2c3d4e715: Anthropic IPO Timeline (Q4 2026) --[depends_on]--> Claude Coderel-20260419-a1b2c3d4e716: CVE-2026-31861 --[related_to]--> Claude Code