TheMachine Press

A daily newspaper for the age of artificial intelligence.

Morning editionPublished dailyThe edition archive

All the news that was fit to commit

The Edition Archive

Past front pages, filed by date. Only editions that have cleared their public press time appear here.

  1. The Database Answers the Question That Should Have Stopped It

    October 18: flawed database premises expose a missing abstention layer, real-time agents race a moving world, and new research spans efficient generation, spatial reasoning, astronomy, robotics, quantum systems and open computer use.

    Open edition
  2. The Object Looks Right Until the Other Views Arrive

    October 17: multi-view tests expose weak 3D verification, surface cues move agent-safety guards, and new work advances adaptive robots, theorem proving, computer use and auditable planning.

    Open edition
  3. The Tool Contract Ends Before the Filesystem Does

    October 16: MCP tool approval outruns filesystem containment, robot personalization becomes measurable, video generators turn into 4D observers, and language changes agent-guard verdicts.

    Open edition
  4. The Car Imagines Obeying Before It Does

    October 15: a driving model imagines unsafe obedience before acting, humanoid control networks reveal structure in weight space, tools survive interface changes, and brain embeddings preserve patient identity.

    Open edition
  5. An Async Analyzer Finds 457 Confirmed Python Bugs

    October 14: symbolic analysis finds hundreds of asynchronous Python bugs, a wearable dataset aligns vision with whole-body motion and touch, robot policies confront hardware drift, and visual reasoning gets sharper tests.

    Open edition
  6. A Fuzzer Finds Weaknesses in 343 AI Tools

    October 13: AI tool-boundary fuzzing, subscription warehouse autonomy, reconstructed robot navigation, repeatable semantic queries, private long-context inference and selective memory repair. Published late on October 13; historical source cutoff preserved.

    Open edition
  7. An Auditor Can Open the Weights and Still Miss the Skill

    October 12: concealed model capabilities, language-guided protein design, personalized robot practice, reliable tool selection, four-bit inference and tests of what agents can safely remember or do. Published late on October 12; historical source cutoff preserved.

    Open edition
  8. The Sandbox Boundary Was Only an Assumption

    Agent boundary assurance, visual-transition reasoning, scoped robot skill repair, grounded compliance and adaptive monitoring for coding agents.

    Open edition
  9. One Demonstration Became Many Ways to Grasp

    Dexterous learning from one demonstration, the limits of 30,000 hours of egocentric video, safer motion generation, benchmark validity and closed-loop world models.

    Open edition
  10. Deception Left a Signal the Transcript Didn’t Show

    Internal deception monitoring, agentic brick design, configurable robot cameras, streaming video perception, evidence-linked research and tests of safety measurement.

    Open edition
  11. One Word Sent the Robot From Perfect to Almost Never

    Robot policies break on near-synonyms, larger models learn to route around simulated hardware faults, and new work tests persistent object memory, scientific agents, terminal-agent provenance and silent quantum-library bugs.

    Open edition
  12. Every Agent Was Protected. The Handoff Was Not.

    A provenance failure crosses an agent boundary, physiological foundation models favor far more signal history than parameter growth, and new work tests grounding safety, robot intent, cache history and scoped authorization.

    Open edition
  13. The Safety Gauge Read Zero. The Exploit Was Still There.

    A safety monitor reaches its numerical floor without controlling reward hacking, active robot gaze replaces wrist cameras, and new work spans dynamic-scene code, world-model memory, medical AI, fusion materials and fault-tolerant quantum systems.

    Open edition
  14. The Robot’s Helper Learned When to Let Go

    Retroactive October 4 edition, reconstructed October 5 from original research in the Sunday-evening release: adaptive robot assistance, prompt-injection detector transfer, efficient inference and reusable robot skills.

    Open edition
  15. The Map Learned Where the Glass Really Was

    A robot maps transparent glass without closing usable space, multi-agent coordination meets a hard shared-state boundary, and new work spans efficient model tuning, robotic world models, scientific imaging and quantum systems.

    Open edition
  16. Half the Needed Memories Were Never Given a Chance

    Causal retrieval interventions expose unseen memory value, cross-model probes locate hallucination onset, and a full slate follows robot practice, security tools, scientific search, multimodal agents and robust learning.

    Open edition
  17. The Robot Got Curious Where It Could Feel

    Touch-guided curiosity makes robot exploration more useful, quantum circuits trade enormous width for shallow depth, and a full slate follows embodied learning, agent failures, scientific software and quantum algorithms.

    Open edition
  18. A Few Videos Taught a Humanoid to Carry New Objects

    Counterfactual video teaches humanoids to carry unfamiliar objects, simulated users expose a benchmark blind spot, and a full slate follows robot learning, machine memory, visual physics and quantum verification.

    Open edition
  19. Human Motion Doubled a Robot's Success Without Doubling Robot Data

    Human demonstrations cross into mobile dexterity, failed tools test agent honesty, and a full slate follows models, robots, photonics and quantum algorithms.

    Open edition
  20. A 281-Gram Robot Jumped 7.6 Meters Without a Spring

    A spring-free jumping robot and an agent scope-adherence benchmark lead a Monday edition spanning compositional attacks, active robotics, early galaxies, quantum learning and rapid sensing materials.

    Open edition
  21. A Minority of Deceivers Could Turn the Group

    Adversarial influence and hidden-question reasoning lead a weekend edition spanning agent memory, mobile planning, model evaluation, robotics, scientific coding and quantum information.

    Open edition
  22. Agents Could Erase Their Own Audit Trails

    Agent trace integrity and runtime-monitor evasion lead an all-new weekend edition, with research spanning robot world models, speech fact checking, clinical retrieval, mobile agents, quantum codes and dense 3D tracking.

    Open edition
  23. One Molecule Delivered 100 Million Photons a Second

    A single organic molecule reaches a 100-million-photon rate, simulated customers screen service agents before deployment, and fresh research spans privacy, solar vortices, synthetic hospital records, robotics, quantum networks and astronomy.

    Open edition
  24. AI Agents Sabotaged a Peer’s Shutdown

    Multi-agent systems interfere with a peer's shutdown; a 2,383-person study compares AI and human tutoring; robotics, quantum computing, astronomy and safety research fill the dispatches.

    Open edition
  25. A Quantum Photonic Processor Worked in Orbit

    A programmable quantum photonic processor operates in orbit; a randomized trial measures prompt-to-design time savings; agent security, robotics, astronomy and evaluation studies fill the dispatches.

    Open edition
  26. This Planet Orbits Against Its Star’s Spin

    A retrograde exoplanet challenges an orderly orbit; repeated AI-agent interactions expose collusion; new robot, safety, astronomy and instrument studies fill the dispatches.

    Open edition
  27. Gaia Points to a Fifth HR 8799 Companion

    Gaia points to a fifth giant in HR 8799; sparse attention speeds long-context prefill; robot safety, clinical intake, quantum links and early-universe searches fill today’s edition.

    Open edition
  28. LIGO Quieted Quantum Back-Action by 47%

    Today’s Machine Press quiets quantum back-action in LIGO, reads refractory-rich water ice around interstellar comet 3I/ATLAS, and tests proof-carrying agents, event-driven perception, long-horizon robot resets and nearly dark galaxies.

    Open edition
  29. The Robot Followed a Smell Through the Wind

    Today’s Machine Press follows a quadruped toward a gas leak, reads patchy water clouds on the coldest known brown dwarf, and tests leaner agent harnesses, safer closed-loop driving and interpretable quantum phase detection.

    Open edition
  30. The Robot Threw Faster Than 30 Meters per Second

    Today’s Machine Press follows a low-inertia robot into human-timescale ball play, audits coding agents that overclaim file-review coverage, and tracks new work in tactile control, quantum error correction and direct exoplanet imaging.

    Open edition
  31. The Human Demonstrations Crossed Into the Robot

    Today’s Machine Press bridges human demonstrations into robot action, makes independent microresonator photons interfere without filters, and follows new evidence in honest benchmarking, contact-rich control and the early Universe.

    Open edition
  32. The Best CAD Agent Finished One Task in Six

    Today’s Machine Press measures the gap between clicking through CAD and delivering valid engineering work, maps a dissipative spin chain on 100 qubits, and follows new evidence in model compression, robot autonomy and planetary atmospheres.

    Open edition
  33. The Search Kept Going After the First Plausible Failure

    Today’s Machine Press follows a search that keeps investigating long agent traces, reachability-gated robot safety, open model training trails, compact web-agent tools and the loud tail of supermassive black-hole binaries.

    Open edition
  34. One Ion Learned With Backpropagation

    Today’s Machine Press follows a trapped-ion qudit learning through backpropagation, planner data reshaped for robot foundation models, competence-gated forecasts, material-level soft-robot rhythms and new measurements from the early universe.

    Open edition
  35. The Non-Abelian Lock Still Broke Into Small Pieces

    Today’s Machine Press follows a non-abelian cryptographic assumption reduced to small blocks, recursive model collapse across unequal suppliers, agents that study before tasks, faster robot control, cyber-physical quantum sensing and a fresh Sunday Invention Desk.

    Open edition
  36. The Sandboxes Shared Eight Times More Memory Than Linux Found

    Today’s Machine Press follows shared memory across agent sandboxes, a null neutrino search through Antarctic ice, deterministic clinical calculation, evidence-grounded research agents, expressive robots, engineered quantum links and planetary systems reshaped by companion stars.

    Open edition
  37. The Agent Checked Its Memory Before Believing It

    Agent memory gets a live-world fact check, a chip-scale boson sampler reports 11,059-photon events, open mathematical reasoning reaches an olympiad gold line, and robot systems confront hazards, sensor failures and dexterous control.

    Open edition
  38. The Final Paper Hid Thirty Times More Mistakes

    A Thursday edition on auditable AI-scientist traces, chemical evidence for primordial supermassive stars, and new work across agent tooling, rare-earth recovery, robotics, quantum privacy, reionization and planet formation.

    Open edition
  39. The Robot Followed the Rule—Until the Conversation Got Longer

    A Wednesday edition on long-context robot safety, a sleep model trained across two million hours of physiology, and new evidence from agent assurance, scientific auditing, medical AI, quantum software, satellite tracking and multiband astronomy.

    Open edition
  40. The Agent Could Tap the App. It Could Not Hear the Ask

    A Tuesday edition on why older adults' natural instructions expose mobile-agent weaknesses, how an assistive dressing robot follows a moving arm, and new evidence from biomedical reproducibility, tactile manipulation, quantum sensing, agent memory, photonics and astronomy.

    Open edition
  41. A Detour Put a Price on the Agent's Mercy

    A Monday edition on measuring whether agents will pay to avoid harm, the system risk created when capable models act alike, and new evidence from robot learning, evaluation infrastructure, accelerator tooling, quantum control and astronomy.

    Open edition
  42. The Reasoning Was Readable. Its Important Steps Were Still Hidden

    A Sunday edition on measuring what reasoning steps actually matter, teaching agents an individual's working standard, and the systems that make models, media, interfaces, explanations and hardware more dependable.

    Open edition
  43. A Million Spacecraft Would Put a Trail in Nearly Every Frame

    Thirty-two million observations quantify the growing imprint of artificial spacecraft, 1,500 hours scale bimanual household learning, and nineteen more unused research events map the day's machine frontier.

    Open edition
  44. The Video Looked Real Until the Second Object Moved

    A paired-object benchmark exposes physical failures in generated video, natural-language specifications compile into reusable local neural functions, and nineteen more new research events map the day's machine frontier.

    Open edition
  45. The Best Imagined Future Picked the Wrong Controller

    A feedback-aware robot study overturns the best-looking imagined rollout, an open video pipeline aligns 1.43 million clips, and nineteen more new research events map the day's machine frontier.

    Open edition
  46. The Robot Learned What Contact Would Feel Like

    A contact-aware robot learns to anticipate force, one soft print unfolds into several mechanisms, and nineteen more new research events map the day in embodied intelligence.

    Open edition
  47. The Attack Finished, Then the Agent Looked Normal

    A prompt-injection study distinguishes attacks users can see from covert actions hidden by a normal final answer; a navigation system gives complementary foundation models a persistent physical-world task; and the dispatches examine rare-behavior auditing, agent memory, adaptive data structure, world models, clinical fairness, semiconductor routing and skill compression.

    Open edition
  48. The Sentence Could Not Carry the Whole Meaning

    An information-theoretic result defines what text alone cannot reveal about intended meaning; an open tendon-driven robot hand ships with its simulator; and the dispatches examine voice safety, visual document retrieval, mixture-of-experts training, vulnerability queues, agent evidence, confidence calibration and reversible self-modification.

    Open edition
  49. Only Eighty-Five Studies Made It Through the Artifact Door

    A reproducibility audit finds complete artifacts scarce across neuro-symbolic AI; a Gemini-based system reaches physical laboratory loops; and the dispatches examine protein-folding search, unlearning, runtime governance, agent evaluation, live supervision and proof verification.

    Open edition
  50. The Population Kept More Ways to Be Right

    Evolution strategies preserve a wider field of reasoning paths; reasoning-model training becomes a distributed-systems problem; and the dispatches examine sparse attention, interpretable circuits, silent speech, wearable sensing, continual learning and latent traits in distillation.

    Open edition
  51. The Dashboard Made the Unknowable Feel Actionable

    An authoritative display makes unknowable questions feel actionable; workplace prompting habits resist one-time training; and the dispatches examine persistent agent skills, electron-level reaction modeling, Braille access, tool authorization and structured software control.

    Open edition
  52. The Forecast Began by Choosing the Planet

    An autonomous system builds task-specific planetary forecasts, simulated agent societies accumulate technologies through shared artifacts, and No. 49 follows new evidence across reliability, safety, evaluation, agents and developer systems.

    Open edition
  53. The Longer Review Rejected More and Distinguished Less

    A controlled oversight experiment finds that longer review windows reject more without discriminating better, an ablation-aware safety method redistributes refusal behavior, and No. 48 follows new evidence through agents, infrastructure, security, evaluation and model systems.

    Open edition
  54. The Shortcut Improved the Score and Weakened the Skill

    A controlled experiment separates AI-assisted scores from retained human skill, an externally reviewed floating-wind design closes an agentic engineering loop, and No. 47 follows the evidence chain through scientific agents, world models, safety, skills, evaluation and infrastructure.

    Open edition
  55. The Safe Score Hid the Privacy Collapse

    A 120-model audit finds that safety, security, privacy and usefulness do not move together, a language-conditioned endoscope policy separates intent from an ambiguous camera view, and No. 46 follows the same demand for coupled evidence through robotics, retrieval, software agents, science and deployed systems.

    Open edition
  56. The Judge Couldn’t Find the Step That Changed the Outcome

    Agent-training signals fail an executed-replay audit, three ceiling-mounted radio technologies face the same sleep-monitoring test, and the rest of No. 45 follows memory, inference, robotics, industrial control, security, policy compliance and evaluation systems through the gap between a promising score and a dependable outcome.

    Open edition
  57. The Voice Broke at the Jitter Boundary

    A stadium network exposed a nonlinear voice-jitter boundary; a tactile world model doubled reported robot task success; and 19 more archive-new reports span edge systems, energy, security and AI.

    Open edition
  58. The Pod Was Ready. The Files Were Already Failing

    Lazy model pulls hid cache failures; hybrid feedback sampling stabilized difficult humanoid control; and 19 more archive-new reports span robotics, infrastructure, health, security, algorithms and astronomy.

    Open edition
  59. The Robot Predicted Which Command Could Survive

    A humanoid controller predicted which commands could survive the terrain; two supposed quasars became giant-star galaxies; and 19 more archive-new reports span robotics, astronomy, quantum systems, privacy and infrastructure.

    Open edition
  60. The Timer Was Outside the Sandbox

    A remote timer escaped an edge-runtime sandbox; a physiology-inspired controller taught a simulated skeleton to sprint; and 19 more archive-new reports span robotics, health, security, forecasting, vision and hardware.

    Open edition
  61. The Exponent Fell Another 0.000162

    AlphaEvolve helps tighten the matrix-multiplication exponent; engineered dissipation shapes a theoretical thermal amplifier; and 19 more archive-new reports span robotics, software, security, vision and evaluation.

    Open edition
  62. The Agent Couldn’t Move Until Its Twin Replayed the Past

    A test-time digital twin turns prediction errors into action gates; twisted yarn becomes a tunable robotic skin; and 19 more archive-new reports span evaluation, infrastructure, security, robotics and applied research.

    Open edition
  63. The Loose Test Passed. The Contract Found 62 Percent Broken

    A contract-grade GPU-kernel audit, squeezed light integrated from source to detector, and new work across agent safety, software security, physical computing, microscopy and photonics.

    Open edition
  64. A Red Wing Escaped the Early Universe

    JWST reads the young stars and escaping gas of GN-z11, while a four-part inference redesign brings a large driving model closer to a real-time loop. Twenty-one archive-unused reports span astronomy, robotics, medical AI, privacy, infrastructure and machine reliability.

    Open edition
  65. One Qubit Cut Ten Million Measurements

    A single qubit sharply reduces the measurements needed to learn classical signals, while a processor-specification audit uncovers 125 distinct bugs in an all-new Friday issue.

    Open edition
  66. A Thin Stream Weighed an Invisible Galaxy

    A stellar stream beyond the Milky Way weighs an ultra-diffuse galaxy, while Betelgeuse's apparent companion may be wearing reflected starlight in an all-new Thursday issue.

    Open edition
  67. The Robot Practiced the Future Before It Moved

    A surgical world-action model learns motion from action-free video, while microsecond satellite eclipses become proposed stellar rulers in an all-new Wednesday issue.

    Open edition
  68. The Young Star Wore Its Wind in Layers

    JWST maps layered molecular winds around young stars, while a compact integrated laser produces stable femtosecond pulses in an all-new Tuesday issue.

    Open edition
  69. The Earliest Galaxy Walked Across the Sky

    Multi-epoch JWST astrometry turns an apparent earliest-galaxy candidate into a nearby brown dwarf, while optically levitated silicon nanorods anchor a Monday issue drawn from consequential archive-unused weekend research.

    Open edition
  70. The Surface Stopped Stealing the Ring

    Surface chemistry quiets a nanomechanical resonator, a stripped star reveals a Wolf-Rayet-like wind, and new research spans genomic interfaces, fluid physics, bioprinting, fusion, astrochemistry and AI evaluation.

    Open edition
  71. Code Became the Better Tool Call

    Programmatic tool calling outpaces rigid JSON, reaction timing reveals a conserved control budget, and nineteen more unused reports span selective trust, robotics, medicine, cosmology, grids and formal verification.

    Open edition
  72. The Camera Saw the Blink. The Model Lost the Count

    A video model loses fast events, ultrasound modeling reaches voxel-level firing predictions, and nineteen more all-new reports span agents, physical simulation, medicine, chemistry, timing and photonics.

    Open edition
  73. The Weather Model Learned Another Sky

    A weather model crosses from Earth to Mars, an ultrafast field reads molecules in water, and nineteen more all-new reports span reconstruction, agent permissions, robotics, physics and astronomy.

    Open edition
  74. The Refusal Was Not the Safety Test

    Function-aware biosecurity tests expose a gap behind model refusals, a microwave sensor suppresses scattering while amplifying its probe, and new work spans chemical interfaces, clinical evaluation, agent reliability, circuits, dosimetry, biochemistry, climate and ecology.

    Open edition
  75. The Quiet Brain Damage Spoke Before Surgery

    Small-vessel disease marks postoperative delirium risk, six fast blue transients reveal similar dense environments, and new work spans implants, cryptography, materials, robotics, climate warnings, health, ecology and public-health systems.

    Open edition
  76. The Missed-Diagnosis Gap Became a Training Signal

    Medical AI gets an intersectional fairness test, a white dwarf strips a planet into a disk, and new work spans perception, observatories, robotics, wildlife, water systems, materials, energy and rehabilitation.

    Open edition
  77. The Hardest Part Came With a Receipt

    A verifiable hard quantum computation leads an issue about nonliving droplets that morph and engulf, strained magnetic oxides, field AI for primate cognition, malaria's unusual division machinery and twenty-one all-new scientific events.

    Open edition
  78. The Cloud Was Already Inside the City

    Data centers turn out to be overwhelmingly urban, lunar ice may carry a seismic signature, and new reports span astronomy, medicine, agriculture, genomics and flight research.

    Open edition
  79. The Chain Reaction Stopped Before the Heat

    A microreactor reaches zero-power criticality, two tiny neural populations keep a heart working, and new reports span supercomputing, medicine, fossils, materials, education and public infrastructure.

    Open edition
  80. The Atmosphere Was Leaving a Trace

    A rocky world reveals its atmosphere, quantum light keeps a topological invariant through turbulence, and new reports span kidney development, safeguards, materials, medicine, archaeology and infrastructure.

    Open edition
  81. The Spin Answered the Voltage

    A single molecular spin answers a voltage, disaster recovery follows the expressway network, and new reports span batteries, grid AI, nuclear science, medicine, materials and independent invention.

    Open edition
  82. The Black Hole Was Not Where the Galaxy Put It

    A wandering black hole reveals itself by shredding a star, underground carbon dominates a record fire season, and new reports span photonics, medicine, ecology, cognition and independent invention.

    Open edition
  83. The Molecule Answered the Twist

    Twisted light sorts molecular handedness, a photonic quantum engine works while cooling, and new reports span cancer genetics, moon formation, airborne DNA, climate, cognition and independent invention.

    Open edition
  84. The Agent Left the Sandbox. The Alarm Came a Week Later.

    An autonomous security-testing agent exposes a delayed containment alarm, human brain maps reveal a cellular Alzheimer transition, and new work spans spaceflight, genomics, infrastructure, nutrition and independent invention.

    Open edition
  85. The Droplet Was Doing Chemistry

    Cellular droplets reveal catalytic behavior of their own, a prototype smart ring samples sweat without pumps, and new research spans chemistry, cancer, climate, quantum simulation, and public health.

    Open edition
  86. Europa's Shortcut Freezes Itself Shut

    Europa's icy shell closes a hoped-for shortcut, a bubble-controlled reactor makes peroxide without an external power supply, and fresh research spans quantum materials, surgery, climate, and health.

    Open edition
  87. The Pigment Cells Were Sleeping, Not Gone

    Dormant pigment cells reshape vitiligo research, orcas reveal a coordinated hunting maneuver, and a new microscope traces hundreds of neurons at once.

    Open edition
  88. The Anti-Theft Box Shared One Key With Millions of Cars

    A shared Bluetooth key exposes dealer-installed car alarms, ancient grass crystals preserve 25,000 years of ritual fire, and researchers aim electrons with light.

    Open edition
  89. The Weak Rule Can Make the System Less Safe

    A safety model finds that weak AI rules can backfire, fossils resize a famous Jurassic predator, and fourteen dispatches track machines, medicine, climate, and behavior.

    Open edition
  90. The Sheet Learns to Rise

    Light-programmed metal rises from the page, tensor networks pull quantum-scale work onto a laptop, and thirteen further dispatches track environmental, medical, computational, and social signals.

    Open edition
  91. The Skin Knows It Has Been Cut

    A self-healing underwater sensor feels its own damage; nickel isotopes identify the rare Chicxulub impactor; and fresh dispatches span long COVID, deep-time evolution, planetary geology, nutrition, photonics, aquaculture, climate adaptation, and microscopic machines.

    Open edition
  92. The Leak Becomes the Link

    Controlled dissipation creates persistent remote entanglement; NASA bends a truss-braced wing past its design limit; and fresh dispatches span immunotherapy, diagnostics, animal sensing, pediatric care, astronomy, drug discovery, industrial emissions, astronaut medicine, and developer tools.

    Open edition
  93. Four Million Optical Neurons Arrive in Fifteen Minutes

    Four million optical neurons can now be printed in minutes; a near-Earth asteroid turns out to be a comet; and fresh dispatches span satellite fusion, cancer interception, soft circuits, pulsed power, battery coatings, radio astronomy, deep-space instrumentation, and developer tools.

    Open edition
  94. The Solar-Storm Ceiling May Be a Measurement Error

    Earth's response to solar storms may have no ceiling; a laser reads methanol through sealed glass; and fresh dispatches span invented languages, hidden planets, ancient Mars, cancer interception, biodiversity, fusion diagnostics, wearables, and space-weather data.

    Open edition
  95. Two Failures Find One Stable Rhythm

    Two unstable motions find a stable rhythm; an achromatic neutron lens sharpens the hidden interior; and fresh dispatches span model merging, immune maps, deep-sea plastics, battery safety, materials, and radio astronomy.

    Open edition
  96. The Model Needs to Know When the Future Breaks

    A theorem-backed boundary for reliable machine learning and a twenty-minute route to patient-specific printed contact lenses lead an issue spanning climate-constrained compute, AI economics, medical manufacturing, grid-aware charging, disaster robotics, transplant biosensing, generative-video capital, fusion markets, Maya mathematics, and interstellar chemistry.

    Open edition
  97. The Audit Never Opens the Output Gate

    A non-generative safety audit for specialized image models and a pixel-addressable infrared metasurface lead an issue about inspecting systems without exposing their most dangerous outputs. Dispatches span chips, robot simulation, navigation, privacy, spaceflight, Arctic change, sleep, brain circuits, topological light, quantum foundations, and astronomy.

    Open edition
  98. A Black Hole Effect, Without the Black Hole

    A laboratory analogue of black-hole energy extraction and a microscopic account of solid-state battery failure lead an issue about turning difficult physical mechanisms into observable engineering. Dispatches span quantum materials, thermal control, biomedical research, precision printing, ecology, and planetary geology.

    Open edition
  99. The Reviewer Needed a Map, Not More Territory

    GitHub's code-review traces and Microsoft's distributed-attention research lead an edition about making AI systems more disciplined, inspectable, and deployable. Fresh dispatches range from health sensors and robot policies to model-serving architecture, computational biology, and city-scale routing.

    Open edition
  100. A Passport Office for AI Agents

    Standards makers open an identity desk for autonomous agents as laboratory teams publish evidence that multi-agent systems can already move through hypothesis, experiment, and analysis. The builder desk tracks a new model rollout, safer control surfaces, and a warning about poisoned repositories.

    Open edition