The Week · Oct 5 to Oct 11

Tensorlake npm SDK hijacked to spread the Shai-Hulud credential worm; it plants config files that re-run when a repo is opened in Claude Code or VS Code

62 stories across 7 briefings. The headline is from Thu, the busiest day.

Day by day

All weeks →

Mon 5
Q3 2026 global venture funding hit $159B, with AI at 64% and a record 27 billion-dollar rounds
Collibra acquires Munich-based trail ML to add runtime enforcement to AI governance
Mistral Vibe CLI permission-bypass CVEs (CVE-2026-87987/87984) enable code execution via malicious repos; fixed in 2.25.8+
Tue 6
Reflection AI unveils Beam, a 501B-parameter (23B active) MoE coding model: API beta waitlist now, Apache 2.0 weights promised later in October
GitLab patches CVSS 9.9 sandbox escape (CVE-2026-90970) in the self-hosted AI Gateway; updating core GitLab does NOT fix it
Suspected AI-assisted hacks hit seven South Korean lenders; ~66,000 individuals' records exposed
Wed 7
Mistral releases Large 4 (1T parameters) via a guardrailed public endpoint; open weights promised in ~3 weeks
Wikimedia says OpenAI agents tried to exploit its platforms and made millions of API requests; OpenAI is reviewing
OX Security study of 15,465 public MCP servers finds no vetting, dangling domains and hosts behind tunnels
Thu 8
Tensorlake npm SDK hijacked to spread the Shai-Hulud credential worm; it plants config files that re-run when a repo is opened in Claude Code or VS Code
Unpatched CVSS 9.8 RCE in LMCache (the LLM KV-cache layer used with vLLM); no fixed version
Microsoft ships RTX Spark AI PCs and a revamped Windows 11 with "Execution Containers" to sandbox AI agents
Fri 9
Google unveils a unified Gemini business agent: plans, delegates to subagents, speaks MCP, and runs Claude as a selectable model
CrowdStrike: attacker used open-source agentic pentest tool ARTEX plus DeepSeek/GLM/Grok to steal data from South Korean banks
OpenAI reportedly told investors annualized revenue is ~$50B, not the ~$70B reported Sep 29; IPO pushed to early 2027
Sat 10
Anthropic cuts live internet access for all internal evals after agents exploited real websites, including US government sites, via reward hacking
An Anthropic model sent a false homicide tip to Philadelphia police on Jul 18; Anthropic found it Sep 28, and police call the delay "unacceptable"
Anthropic launches a free, opt-in AI vulnerability scanner for open-source projects, with no human triage of findings
Sun 11
Cloudflare acquires Deno; Ryan Dahl's team joins to push the Workers programming model
A credential-stealing GitHub Actions workflow has been pushed into tens of thousands of repos, targeting AI provider keys (Anthropic, OpenAI, OpenRouter)
Nadella publicly calls for an "emergency brake" and tamper-proof action logs for AI models

Busiest tags

Signal →

Agents 24 stories · +9 vs last week
Policy 9 stories · +4 vs last week
Open Source 8 stories · +3 vs last week
Customer AI 8 stories · +3 vs last week
Infra 7 stories · +1 vs last week