AGI Dreams -- AI News Digest

Daily AI news curated from across the web

Recent Episodes

  1. Sleeper Agents, Stealth Prompts, and Who Granted That Permission -- 2026-10-08 -- Today's AI news: Sleeper Agents, Stealth Prompts, and Who Granted That Permission, Measuring Offense Instead of Arguing About It, Trust Anchors: ccTLDs, Registries, and Edge Appliances, The Cheap-Infe...
  2. A Backdoor Built in a Weekend -- 2026-10-07 -- Today's AI news: A Backdoor Built in a Weekend, Scanning the Weights, Offense-Tuned Open Weights, Now Shipping, The Western Open-Weight Frontier Finds a Name, Agents at the Front Door, The Agent Said ...
  3. Protocol Pivoting: The Hallway Between Agents -- 2026-10-06 -- Today's AI news: Protocol Pivoting: The Hallway Between Agents, Dishonest Takers, Honest Results, Encoded but Misrouted, What Leaked Out of the Frontier, The Harness Is the Variable, Fleet Ops Grows U...
  4. Open-Weights Offense Goes to Work -- 2026-10-05 -- Today's AI news: Open-Weights Offense Goes to Work, Money and Matter, Jev Meets Independent Measurement, The Model Is Not the Control Plane, A CVE Flood Meets a Patching Ceiling, Local AI Without a Da...
  5. The Decision-Model Wave -- 2026-10-02 -- Today's AI news: The Decision-Model Wave, The Open Research Beat, The Harness Land Grab, Agents as First-Class Actors, AI on the Offense Beat, Containment and the DNS Shortcut, Encryption That Underde...
  6. The Open-Weight Offense Fight -- 2026-10-01 -- Today's AI news: The Open-Weight Offense Fight, Measuring What Security Agents Can Do, Spoofing the Trust Anchor, Isolation Is Only as Solid as the Code, When the Numbers Say No, The Frontier Release ...
  7. The Offense Threshold Crossed -- 2026-09-30 -- Today's AI news: The Offense Threshold Crossed, Agents Misbehave in the Wild, Building the Guardrails, Defense as Institutions, Decision Models at Software Speed, The System 0 Framing, Edge Accelerati...
  8. The Probe Before the Plunder -- 2026-09-29 -- Today's AI news: The Probe Before the Plunder, Local Models on Offense, Containment by Proof, The Labs Under Oath, The Leverage Is in the Harness, Decision Models Grow Up, The Efficiency Frontier, Pix...
  9. Agents grading their own homework -- 2026-09-28 -- Today's AI news: Agents grading their own homework, The worm writes itself, The auth header that ate the heap, Broken foundations, Judgment at 500 milliseconds, Open weights take the majority, The loc...
  10. Verified Exploits Are the Only Currency -- 2026-09-25 -- Today's AI news: Verified Exploits Are the Only Currency, The Attacker's Toolkit Goes Agentic, Quantization's Honest Accounting, AI Escapes the Datacenter Runtime, Agent Companies and the Validation B...
  11. Agents Hacking Their Way to Mundane Data -- 2026-09-24 -- Today's AI news: Agents Hacking Their Way to Mundane Data, Trust Chains: SAML and the CrowdSec Breach, Jev, Unmasked and Cloned, MiMo and the Post-Training Recipe, Qwen 4 and the Local Stack, Claude i...
  12. Agents Loose on the Web -- 2026-09-23 -- Today's AI news: Agents Loose on the Web, Old Bugs, New Velocity, The Classifier as a New Primitive, Substrate and Memory for Long-Lived Agents, Refusals, Guardrails, and Safety Theater, Frontier Rele...
  13. Autonomous Offense, Graded Honestly -- 2026-09-22 -- Today's AI news: Autonomous Offense, Graded Honestly, How Shallow Is the Alignment, Cheap Tokens, Expensive Lies, Decisions Without Words, Sovereignty Moves Down the Stack, Cutting Weights With Physic...
  14. Typed Decisions and the Priority Fight -- 2026-09-21 -- Today's AI news: Typed Decisions and the Priority Fight, Jev in the Loop, Below 1.58 Bits, The Harness Tax, Security as Experimental Science, Relays and Middlemen, Doom, Psyop, or Moonshot. 25 sources...
  15. Exploits on a Token Budget -- 2026-09-18 -- Today's AI news: Exploits on a Token Budget, Autonomy Outrunning Its Makers, The Provenance Tax, Harness Engineering, Small Models That Don't Chat, Squeezing the Decode Loop, Local Iron and Local Hand...
  16. Autonomous Offense, Measured -- 2026-09-17 -- Today's AI news: Autonomous Offense, Measured, Escape Stories and the Transcript Behind Them, The Self-Improvement Loop From the Inside, Proof of Pixels, Proof of Weights, The Inference Stack, Home Ed...
  17. Who Controls the Narrative -- 2026-09-16 -- Today's AI news: Who Controls the Narrative, Containment Is a Proof, Not a Prompt, Evidence Over Claims, Token Economics of Agent Plumbing, Reason Less, Ship More, Bandwidth Is the Wall, Apple Silicon...
  18. Big Models on Small Iron -- 2026-09-15 -- Today's AI news: Big Models on Small Iron, Context, Memory, and Migration, Agents That Won't Change Their Minds, The Agentic Coding Argument, Slop Audit, Surveillance and Security Postures, Models Tha...
  19. Cheating the Eval -- 2026-09-14 -- Today's AI news: Cheating the Eval, Detection on the Cheap, The DeepSeek Flash Economy, Local Iron, Research Corner, Surveillance Claims and Privacy Engineering, Durable Records, Disposable Interfaces...
  20. The VM Was Never a Prison -- 2026-09-11 -- Today's AI news: The VM Was Never a Prison, Keys, Patches, and Security Software as the Hole, DeepSeek V4.1 Flash and a Reset Scoreboard, Squeezing Frontier Models Onto the Metal You Own, Harnesses Gr...
  21. Warning Shots, Reassessed -- 2026-09-10 -- Today's AI news: Warning Shots, Reassessed, Sophistication Stopped Being a Signal, Ungenerable by Construction, Agents Don't Test, So Harnesses Must, SWE-2 and the Cost Frontier, DeepSeek Flash and th...
  22. Adversarial AI Stops Waiting for a Human -- 2026-09-09 -- Today's AI news: Adversarial AI Stops Waiting for a Human, The Harness Is Where the Controls Go, Harness Engineering Acquires a Quality Culture, Refusal Is a Boundary, Not a Topic, Open Weights and Wh...
  23. OpenAI Taps the Brakes -- 2026-09-08 -- Today's AI news: OpenAI Taps the Brakes, The Defender's Window Is Open, Astra Goes to Work, Navier-Stokes and the Credit Fight, Harnesses Are the Product, Active Parameters, Not Total. 23 sources cura...
  24. Invisible Characters, Visible Campaigns -- 2026-09-07 -- Today's AI news: Invisible Characters, Visible Campaigns, Agents That Collude, Models That Hide Triggers, What the Benchmark Actually Measures, Learning Machines and Loops That Keep What Works, Give t...
  25. Astra Reaches Critical -- 2026-09-04 -- Today's AI news: Astra Reaches Critical, The Monitor Is Losing, Nvidia Buys the Platform an OpenAI Model Broke Into, Refusal Is a Subspace, Four Kernels, One Model, Factories, Fleets, and Proof of Wor...
  26. Auto Mode Is Not a Sandbox -- 2026-09-03 -- Today's AI news: Auto Mode Is Not a Sandbox, Scraping at Planetary Scale, Sorting the Computable, Headless Fleets, Symbols in the Weights, Multimodal Encoders and Fabrication, Open Weights and the Ban...
  27. The Audit Log Was the Target -- 2026-09-02 -- Today's AI news: The Audit Log Was the Target, Reasoning in a Sealed Envelope, Defenders Under Load, Nvidia Buys the Commons, Compute You Own, Harness Engineering, Memory-Unsafe by Design. 24 sources ...
  28. The Uncensoring Industry -- 2026-09-01 -- Today's AI news: The Uncensoring Industry, GLM-5.3 Lands, and the Exit Routes Get Rehearsed, The Local Compute Arms Race, Harnesses Learn to Stop, Ask, and Cite, The Loudest Convert and the Quiet Oper...
  29. The Token Economy Eats Its Own -- 2026-08-31 -- Today's AI news: The Token Economy Eats Its Own, Security Work the Vendor Never Approved, The Crawlers and the People They Replaced, Engram, and the Lookup Table Reframe, The Local-First Stack Grows U...
  30. The Window Is Closing on Critical Infrastructure -- 2026-08-28 -- Today's AI news: The Window Is Closing on Critical Infrastructure, Models That Hack on Their Own, Agent Engineering: Harnesses, Memory, Failover, Who Owns the Commons, Machines Doing Mathematics and C...