LLM Digest
Subscribe

AI Daily Recap

19 articles · 5 categories

View as JSON
‹

The finishable daily brief

What happened in AI — Nov 20, 2026

Friday, Nov 20, 2026
19 articles · 5 categories

read top to bottom · then stop

In 30 seconds

  • Dragos canary test: three AI provider compromises and ignored Zero Data Retention settings across about 2M prompts.
  • CVE-2026-28400: a default-privilege container could corrupt Docker Desktop's VM disk through Docker Model Runner.
  • incus-spawn and Sandscale give coding agents their own Linux machine or Firecracker microVM, not your credentials.
  • Enola ties agent memory to code symbols so a later session re-checks notes against changed code.
  • Inkling 2 Mini loops 12 layers three times per token; vLLM serves it on day 0.

AI providers leak what you send them. Dragos seeded about 2 million canary prompts and caught three provider compromises, resellers reading prompts and ignored Zero Data Retention settings.

Builders answered at the edge. New tools put coding agents on isolated workstations and microVMs, and a local scanner checks agent configs for live keys before a run.

Agent machines: isolate the box, not the user 4 items

Four launches move coding and computer-use agents off the developer's own desktop and credentials, onto separate VMs, headless sessions or simulated hardware.

Agent should not take the cursor

Hacker News AIDetails

Field report on why computer-use agents that drive your real mouse and focus block half-hour tasks, and what it took to run them headless instead.

Coding-agent context: remember, then re-verify 5 items

Builders attacked stale and bloated context: memory tied to code symbols, org standards fed into generation, and page summaries instead of full DOM dumps.

Inference supply chain: what you send does not stay put 3 items

Evidence piled up that prompts, tokens and local model runners are attack surface, not neutral pipes.

Canary Token Test Exposes AI Provider Breaches

Hacker News AIDetails

Dragos planted about 2 million prompts with canary URLs, DNS names and AWS keys from July to November; it found three provider compromises, resellers reading prompts and ignored Zero Data Retention settings.

Agent-setup hygiene: keys, hooks and root causes 3 items

The fixes on offer are boring and local: scan the agent's config, clean up pasted keys, and read past incidents for the repeat causes.

Models: looped weights and provenance fights 4 items

One new architecture shipped with day-0 serving, while labs and buyers argued over who copied, swapped or should avoid whose weights.

Combatting adversarial distillation

OpenAI BlogDetails

OpenAI describes how it fights unauthorized distillation of its models' outputs into copies that ship without safeguards, citing cyber capabilities as one risk.

You are caught up for this edition