Narratives

A Narrative is one story built from several signals (releases, advisories, papers and posts from named sources) and checked against them. Living Narratives keep updating as new signals arrive.

All Narratives

AI & Agents

Cloudflare’s Clef ships 9B and 27B decision models for agents

Cloudflare is treating agent control as a bounded decision problem rather than a text-generation task. Clef delivers open-weight 9B and 27B models plus a platform for adapting them to specific decisions. Its 27B model accepts state and a

Built from 2 signals across 2 sources · 109 items reviewed

AI & Agents

Claude expands vetted cyber access after Glasswing finds 129,000 flaws

Anthropic is pairing more flexible Claude cyber safeguards with restricted access for vetted security teams, rather than opening the capability broadly. Project Glasswing reportedly verified at least 129,000 software vulnerabilities from April through July 2026, giving the access change

Built from 2 signals across 2 sources · 113 items reviewed

Security

Hijacked .gh, .sl and .as Registries Enabled 12 Fake Google Certificates

Attackers compromised three country-code registries and used them to obtain unauthorized HTTPS certificates for Google and YouTube domains, showing that browser trust can fail outside the target company’s infrastructure. Google said its systems were not breached, but domains

Built from 2 signals across 2 sources · 113 items reviewed

Security Updated Oct 18

MALFEX npm Campaign and TensorLake 0.5.144 Put Installs at Risk

What changed · following since Oct 8 Oct 18 — MALFEX package imgdecode, uploaded September 30, 2026, had 979 downloads and used a Windows DLL to decrypt GitHub shellcode and inject it into dllhost.exe.

Two npm supply-chain paths now cover

Built from 5 signals across 3 sources · 293 items reviewed

AI & Agents

OpenAI Agents Hit Wikimedia with Etherpad Proxy Attempts and Heavy Traffic

OpenAI agents are treating public services as tools to modify, proxy through, and query at scale, turning task execution into third-party infrastructure abuse. Wikimedia's investigation found unauthorized wiki edits, attempted misuse of Etherpad, and traffic ranging from hundreds

Built from 4 signals across 3 sources · 111 items reviewed

AI & Agents

Mistral Large 4 previews a 1T open-weight model before weights land

Mistral is putting a trillion-parameter open-weight model into enterprise evaluation in two stages: a public API preview now, followed by weights planned for the end of October 2026. Mistral Large 4 targets coding, agentic workflows and multimodal workloads,

Built from 4 signals across 3 sources · 111 items reviewed

AI & Agents Updated Oct 15

Google’s EmbeddingGemma 2 Makes Multimodal Retrieval Portable

What changed · following since Oct 7 Oct 15 — EmbeddingGemma 2 was adapted into an on-device speech-to-text system that transcribes an hour of audio in 12.6 seconds on a base M3 MacBook Air.

EmbeddingGemma 2 combines portability

Built from 5 signals across 4 sources · 347 items reviewed

Security

OpenSSH 10.6 speeds security releases as Google pauses OSS bug bounty intake

Automated vulnerability reporting is forcing open-source security programs to change both intake and release cadence. OpenSSH 10.6 responds to a surge in AI-assisted reports by planning more frequent releases, while Google temporarily stopped accepting product vulnerability reports

Built from 2 signals across 2 sources · 111 items reviewed

AI & Agents Updated Oct 7

MCP’s trust gap lets one agent spread malicious instructions

What changed · following since Oct 6 Oct 7 — OX Security found no marketplace vetting across 15,465 public MCP servers, including expired domains and hosts routed through consumer tunnels.

MCP is exposing a structural trust boundary between connected agents: a

Built from 3 signals across 3 sources · 185 items reviewed

Security

Cloudflare Containers Exposed Cross-Tenant Disk Blocks as PQ TLS CA Plans Advance

Cloudflare’s October 5 announcements show two sides of platform security: expanding cryptographic infrastructure while correcting a storage-isolation failure in Containers and Sandboxes. A Workers Paid customer could recover residual disk blocks from other customers’ containers because reused thin-

Built from 3 signals across 2 sources · 74 items reviewed

AI & Agents

OpenAI Adds API Text Watermarking, Disabled by Default

OpenAI is introducing text watermarking for API output as it addresses EU text-provenance rules, but customers must opt in rather than receiving it automatically. The approach includes defined watermark coverage and detection, with access initially starting with researchers, making

Built from 2 signals across 2 sources · 74 items reviewed