Morning Brief: Thursday, August 13

Sixty-six feeds. Two weeks. 4,603 items reduced to what follows. (what we track, how we crawl, subscribe)

Thursday's dominant thread is state-authorized offensive cyber turning explicit: the Trump administration is publicly enlisting private companies to hack foreign cybercrime groups the day after China-linked actors are reported to have run the first "autonomous" AI-driven cyberattack on Taiwan.

Read together those two Slashdot items form the two ends of the same year: capability (AI can now run an offensive campaign end-to-end) and authorization (states are willing to outsource retaliation to the private sector by name). Everything else on the wire today — Grok 4.6's AINews synthesis, the Codex Desktop landing on Linux, another claude-code point release — is the AI-product cadence continuing underneath the security posture shift.

Top (5-7 min)

Trump Administration Enlists Private Companies To Hack Foreign Cybercrime Groups: Slashdot, 2026-08-13. State-sanctioned offensive cyber outsourced to private contractors, by name. The Letters-of-Marque frame that has floated around policy circles for a decade now has an actual administration attached.
China-Linked Hackers Used AI To Run First-Ever 'Autonomous' Cyberattack On Taiwan: Slashdot, 2026-08-12. Reporting frames it as the first end-to-end AI-run offensive operation, not just AI-assisted tooling. Pairs directly with the Trump-administration item above as the capability side of a story whose authorization side landed 24 hours later.
AINews: SpaceXAI Grok 4.6 and Grok @Bot: Latent Space, 2026-08-13. Thursday synthesis of the Grok 4.6 release from Wednesday plus the @Bot rollout. Latent Space adopting the "SpaceXAI" framing is worth noting — the parent-brand relabel is quietly propagating.
ChatGPT Desktop (Codex Desktop) for Linux: HN → OpenAI, 2026-08-13. Codex Desktop shipping on Linux closes the platform matrix and makes the "coding agent as first-class desktop app" story cross-OS. Follows Tuesday's TNS piece on the same launch reaching Linux.
Prompt Injections for Defense: Schneier, 2026-08-12. Yesterday's post still leading the security thread: prompt injection reframed as a defensive tool. Reads differently after today's private-cyber-contracting news — same category of blurred lines between offense and defense.
Tracking down the 16-year-old WAL-reset SQLite bug: HN → Tailscale, 2026-08-12. Tailscale post co-timed with Antithesis's own writeup. A 16-year-old data-loss window in one of the most-deployed embedded DBs — and it took a deterministic-simulation testing shop to actually surface it.
How kids feel about AI, in their own words: MIT Tech Review, 2026-08-13. Direct-testimony piece rather than a survey summary. Companion to yesterday's "godmother of AI" quote about student motivation risk.

Themes this week

State-authorized offensive cyber: Slashdot: Trump admin enlists private companies to hack cybercrime groups (Thu), Slashdot: China-linked hackers use AI for first "autonomous" attack on Taiwan (Wed), Cloudflare: DDoS Threat Report H1 2026 (1 Tbps attacks) (Tue), HN: Someone is running mass vulnerability scans, spoofing AI bots like ClaudeBot (Wed).
Grok 4.6 rollout: Latent Space: SpaceXAI Grok 4.6 and Grok @Bot (Thu), HN → xAI: Grok 4.6 (Wed), Vercel: Grok 4.6 on AI Gateway (Wed), TNS: SpaceXAI trained Grok 4.6 on something most AI labs throw away (Wed).
SQLite WAL-reset bug (16 years buried): HN → Tailscale: Tracking down the 16-year-old WAL-reset SQLite bug (Wed), HN → Antithesis: Breaking the WAL (Wed).
Reasoning-trace extraction (carryover): Latent Space: How to steal a Reasoning Trace (Wed), SW: Stealing Reasoning Traces (Tue), WIRED: A New Trick Reveals AI Models' Inner Thoughts (Tue).
Coding-agent desktops + release cadence: HN → OpenAI: ChatGPT Desktop (Codex Desktop) for Linux (Thu), TNS: OpenAI ChatGPT/Codex desktop app on Linux (Tue), claude-code v2.1.231 (Thu), claude-code v2.1.229 (Wed), HN → Zed: Introducing Delta (Wed).
Agent substrate: Postgres, sandboxes, connectors (carryover): Databricks: Electric joins Databricks (WASM Postgres) (Tue), Neon: Neon Functions — backend logic next to your data (Wed), Vercel: Set up coding agents in one command with AI Gateway (Wed), ClickHouse: iFood's agentic security platform on ClickHouse (Wed).
Epistemic decay: model collapse + watermarks (carryover): Pluralistic: Model collapse (Wed), SW: No lossless transformations of natural-language text (Tue), TNS: Anthropic watermark survives copy-paste, not dev workflow (Tue).
AI + kids/education: MIT TR: How kids feel about AI, in their own words (Thu), Slashdot: Godmother of AI on student motivation risk (Wed).

Scan (15 min)

Thursday feeds
- Trump Administration Enlists Private Companies To Hack Foreign Cybercrime Groups, Slashdot, 08-13
- Scientists Create Largest 2D Map of the Universe, Slashdot, 08-13
- How Social Media Spurred a Refugee Crisis Between Spain and Morocco, Slashdot, 08-13
- AINews: SpaceXAI Grok 4.6 and Grok @Bot, Latent Space, 08-13
- ChatGPT Desktop (Codex Desktop) for Linux, HN → OpenAI, 08-13
- How kids feel about AI, in their own words, MIT Tech Review, 08-13
- LWN.net Weekly Edition for August 13, 2026, LWN, 08-13
- claude-code v2.1.231, GitHub releases, 08-13
- The Chernobyl NPP and Forty Years of Robots for High Radiation Environments, Hackaday, 08-13
Wednesday carryover — state cyber + AI-attack
- China-Linked Hackers Used AI To Run First-Ever "Autonomous" Cyberattack On Taiwan, Slashdot, 08-12
- Someone is running mass vulnerability scans, spoofing AI bots like ClaudeBot, HN → knownagents, 08-12
- DEF CON Crowd Suspected In Fake-Hotspot Attack On Delta Flight, Slashdot, 08-12
- German Advocacy Group Lodges Criminal Complaint Over Meta AI Glasses, Slashdot, 08-12
Wednesday carryover — Grok 4.6 + coding agents
- Grok 4.6, HN → xAI, 08-12
- Grok 4.6 now available on AI Gateway, Vercel, 08-12
- SpaceXAI trained Grok 4.6 on something most AI labs throw away, The New Stack, 08-12
- Delta, HN → Zed, 08-12
- Set up coding agents in one command with AI Gateway, Vercel, 08-12
- "Issue tracking is dead": how the pull request became the SDLC bottleneck, The New Stack, 08-12
- Anthropic's Chrome extension is now a Cowork session, The New Stack, 08-12
- Anthropic gave agents the ability to dream. Then developers woke up., The New Stack, 08-12
- Code that passes every test can still break the next AI agent that touches it, The New Stack, 08-12
- Your contributors are AI-first now. Is your project?, GitHub Blog, 08-12
Wednesday carryover — SQLite WAL-reset
- Tracking down the 16-year-old WAL-reset SQLite bug, HN → Tailscale, 08-12
- Breaking the WAL, HN → Antithesis, 08-12
Wednesday carryover — models, data
- DeepSeek V4 Pro 0813, HN → OpenRouter, 08-12
- DeepSeek V4 Pro now runs updated weights on AI Gateway, Vercel, 08-12
- Qwen3.8-2.4T, HN → Hugging Face, 08-12
- LFM2.5-VL-3B for Better and Faster Vision Capabilities for the Edge, Hugging Face Blog, 08-12
- OlmoEarth embeddings: custom embedding exports for downstream analysis, Hugging Face Blog, 08-12
- Recall is the bottleneck for parametric factuality, Google Research, 08-12
- MindTopo reveals VLMs' spatial reasoning abilities, Microsoft Research, 08-12
- How a major freight railroad scaled pipeline creation with Genie Code, Databricks, 08-12
- How Amtrak is building the data backbone for its largest transformation in over 50 years, Databricks, 08-12
- Neon Functions: backend logic next to your data, Neon, 08-12
- How iFood built its agentic security platform on ClickHouse Cloud, ClickHouse, 08-12
Wednesday carryover — culture + politics
- Twitch is Mining Peoples' Streams to Train Amazon's AI, 404 Media, 08-12
- AI Generated 3D Models Flood Market, But Almost No One Is Buying Them, 404 Media, 08-12
- Meta 'Pervert Glasses' Used to Harass Women, 404 Media, 08-12
- The Government is Monitoring Anti-Flock TikTok and Instagram Accounts, 404 Media, 08-12
- Big Tech Wants to Harvest Your Thoughts, Slashdot, 08-12
- Godmother of AI on the biggest AI risk in schools, Slashdot, 08-12
- US Tries to Override New York Gambling Laws, Orders Kalshi to Keep Operating, Slashdot, 08-12
- Facebook is paying controversial creators to produce rage-bait, HN → ABC, 08-12
Wednesday carryover — alignment + research
- AI swarms are starting to pose indirect takeover risk, Alignment Forum, 08-12
- An anytime algorithm for mixing the computable measures, Alignment Forum, 08-12
- I wrote an AI textbook — how long until AI can do it better?, Interconnects, 08-12
- Specula: Scaling formal specifications for autonomous model checking of system code, Murat Demirbas, 08-12
- Local State in Livelymerge, Ink & Switch, 08-12
- Learning contact representations in real-world clutter for universal robotic grasping, Nature Machine Intelligence, 08-12
Wednesday carryover — corp + funding
- How RingCentral builds AI-native work from engineering to ops, OpenAI, 08-12
- From assistance to execution: How enterprises put AI to work, OpenAI, 08-12
- Lovable raises $400M Series C, HN → Lovable, 08-12
- Launch HN: Discovered Materials (YC P26) — AI agents to discover new materials, HN, 08-12
- Pontifications: Lingering Resentment Overhangs SPEEA Vote, Leeham News, 08-12
- Airbus A220's Performance Drives BermudAir's Fleet Growth Plan, Leeham News, 08-12
LWN + systems
- Security updates for Wednesday, LWN, 08-12
- QEMU 11.1 released, LWN, 08-12
- A look at CrossPoint e-reader firmware, LWN, 08-12
- Block-layer error injection, LWN, 08-12
Hackaday + hardware
- The Chernobyl NPP and Forty Years of Robots for High Radiation Environments, Hackaday, 08-13
- Have Scientists Stuck the Landing on the Glueball Discovery?, Hackaday, 08-13
- Read a USB Logitech Racing Wheel on an ESP32-S3, Hackaday, 08-12
- The Agony of Packaging Linux Apps, Hackaday, 08-12

Tail

Two Slashdot items are the same story: Wednesday's China-linked "autonomous" AI cyberattack on Taiwan and Thursday's Trump-administration private-cyber-contracting announcement are the capability and authorization sides of a single arc: states can now credibly outsource offensive operations to private firms because AI has made those operations executable end-to-end. The Letters-of-Marque analogy that has been academic for a decade is now the actual news copy.
Grok 4.6 pipeline is complete inside 48 hours: Wednesday: xAI ships. Same day: Vercel AI Gateway lists it, TNS writes the training-data angle. Thursday: Latent Space AINews synthesis. The rollout playbook for a frontier release is now a two-day cadence — release, distribute, contextualize.
The SQLite WAL-reset bug proves DST's value proposition: A 16-year-old data-loss window in one of the most-tested embedded databases on the planet, surfaced by Antithesis's deterministic simulator and independently found by Tailscale. Two of the strongest cases-in-point for deterministic simulation testing landing the same day.
Codex Desktop for Linux closes the coding-agent platform matrix: With Codex Desktop on Linux and claude-code releasing every 1-2 days, the coding-agent-as-desktop-app is now a fully cross-OS, weekly-cadence product category. Zed's Delta on the same wire Wednesday makes it a three-vendor beat.

Feed silences (>72h since last item)

Sources that publish frequently but have gone quiet:

InfoWorld (43 days) — last item 2026-07-01.
Lilian Weng (40 days) — last item 2026-07-04.
Charity Majors (36 days) — last item 2026-07-08.
Kenneth Payne (34 days) — last item 2026-07-10.
Andrej Bauer (33 days) — last item 2026-07-11.
Julia Evans (23 days) — last item 2026-07-21.
Stephen Wolfram (23 days) — last item 2026-07-21.
Fly.io (20 days) — last item 2026-07-24.
Antithesis (17 days) — last item 2026-07-27 (broke silence Wed with WAL-reset writeup — feed hasn't caught up).
METR (16 days) — last item 2026-07-28.
The Markup (16 days) — last item 2026-07-28.
Hillel Wayne (15 days) — last item 2026-07-29.
Marc Brooker (15 days) — last item 2026-07-29.
Grafana Labs (13 days) — last item 2026-07-31.
deepmind-blog (12 days) — last item 2026-08-01.
Murat Demirbas (11 days) — last item 2026-08-02.

Build provenance