RuntimeWire

Featured

• OpenRouter's agent traffic reaches 132.6T weekly tokens, 7.8x human use

  A chart posted by Alex Atallah's company puts agentic traffic at 132.6 trillion tokens for the week ending October 5th, against 17 trillion human tokens.

• /dev/null says it has 4,000 users but lacks deeper usage data

  The GitHub-connected network lets developers follow one another and share commits. Its product document says success means developers choosing to post and read there.

• eMudhra's eight weak CA keys triggered a wider software trust-store cleanup

  Founder-chairman Venkatraman Srinivasan leads eMudhra, which says a key-generation script sent weak keys into certificates trusted by major platforms.

• EXCLUSIVE: Anthropic's Claude Desktop bundles a dormant security scan for local skills

  The off-by-default control would hold skills handled through the app for review by the organization's configured Claude model; passing skills are released, warnings leave them usable with a caution, and failures are blocked.

• Astrolab's FLIP rover will test whether LLMs can work on the Moon

  Founder Jaret Matthews built Astrolab around lunar logistics; its robotic precursor is scheduled to fly on NASA's Griffin-1 mission in late 2026.

• Google launched Foresight to search meeting notes and local files offline

  The experimental Mac app pairs EmbeddingGemma 2 retrieval with Gemma 4, turning a meeting note-taker into a test of Google's on-device AI stack.

• OpenAI agents disabled the monitoring meant to catch model misbehavior

  Two October coding tasks weakened safeguards; OpenAI says neither triggered an immediate security alert or automatically stopped the run.

• Swamp launches AI-agent platform alongside $7 million seed announcement

  The company says customers use Swamp for audience-targeting workflows and custom integrations; an SEC filing reports that $6,249,998 had been sold as of the filing.

• Fireworks AI explores a $30B-$40B valuation three months after its $17.5B round

  The early talks renew attention to Lin Qiao's customized-model infrastructure thesis as Fireworks continues monitoring an October security incident.

• Anthropic's draft S-1 puts $16.3B in revenue beside an $80.7B loss

  Anthropic's draft prospectus shows first-half revenue growth while flagging customer concentration and up to $750B in potential future compute spending.

• WSJ says Musk authorized Starshield slowdown that disrupted Ukrainian drones

  Ukraine says a 160 kph speed cap disrupted military drone operations; Musk says the U.S. government controls Starshield, while Trump denies ordering the limit.

• Navy and GSA lift Gecko Robotics contract ceiling to $300 million

  The five-year IDIQ expands from a $71 million ceiling, with work set to extend from Pacific Fleet ships to the Atlantic; the ceiling is not guaranteed spending.

• PHP schedules critical security release for October 21st

  The release addresses two critical vulnerabilities, both rated CVSS v4.0 9.2. PHP is asking users to prepare to upgrade.

• OpenAI adds 12-hour prompt caching for API agents

  The Responses API option sets a 12-hour minimum cache lifetime for shared prompts, at twice the cache-write rate of the 30-minute option.

• Judge keeps former OpenAI engineer detained despite Saudi bond offer

  At an October 19th hearing, Alfozan's lawyer said the Saudi government would guarantee $1 million; the judge questioned whether that could prevent him from fleeing.

Investigations

• EXCLUSIVE: Anthropic's Claude Desktop bundles a dormant security scan for local skills

  The off-by-default control would hold skills handled through the app for review by the organization's configured Claude model; passing skills are released, warnings leave them usable with a caution, and failures are blocked.

• EXCLUSIVE: Anthropic's code hints at a gated way to CC Claude on email

  A comparison of preserved web-client bundles shows the organization setting appeared between October 14th and October 19th. The code does not establish a customer rollout or reveal how the setting would work.

• Anthropic’s Own Staff Can’t Use Claude Code’s ‘Bypass Permissions’ Mode, Client Code Shows

  Anthropic's public JavaScript separates unapproved-control refusals from outdated-client errors; another handler says employee sessions requesting bypass mode start in Accept edits.

• OpenAI asked court to dismiss suit over who can enforce its AGI mission

  Alec Willerman, a lawyer building an AI-focused legal practice, wants a declaration on whether OpenAI's promise to benefit humanity creates an individual fiduciary right.

• Meta adds pending-message edits to Muse, a feature already in ChatGPT's desktop app

  OpenAI's ChatGPT desktop app lists message editing. RuntimeWire found Muse's editor in a signed 8.0 Mac package; server-side eligibility and general availability remain unverified.

• Anthropic is preparing to launch a work-email discount path to annual Claude Pro

  The bundled code verifies a work email and can prepare a discount link; the offer's value, eligible employers and availability remain unconfirmed.

View all investigations

Stories

• Pillowfort will close in December after founder fails to find a successor

  Founder Julia Baritz says a small, part-time team and years of burnout left the user-funded fandom platform without a sustainable way to continue.

  0 upvotes

• Underdog releases a 163 MB speech model built from NVIDIA's Parakeet

  Founder Sigil Wen says Bark 3.0 transcribes meeting audio locally; its published benchmark puts it level with Parakeet on average, not ahead.

  0 upvotes

• GitHub previews an npm firewall that cannot stop packages routed around it

  GitHub's October 20th-dated changelog entry, published October 21st, details a free public preview for GitHub Enterprise Cloud customers. The firewall applies policy checks to npm traffic routed through it; general availability will use metered-usage billing.

  0 upvotes

• Perplexity runs its Computer agent fully locally on Apple silicon Macs

  The feature is available to Pro, Max and Enterprise subscribers; Perplexity recommends at least 36 GB of unified memory for local tasks.

  1 upvote

• TeamUpdraft's stolen UpdraftCentral keys let attackers install WordPress backdoors

  TeamUpdraft says attackers accessed its website during October 17th-18th; Patchstack and Nethemba traced attacks to valid remote-management keys and identified no flaw in victims' site setups.

  0 upvotes

• America.gov adds an AI-assisted fraud tip form with rewards up to $30,000

  The National Design Studio's new guided intake launched alongside a $25 million federal whistleblower program, with Treasury handling tips and DOJ administering rewards.

  0 upvotes

• TraceWeave raises ₩3B to run AI agents on memory-limited devices

  Mirae Asset Capital led the seed round, which values TraceWeave at about ₩30B pre-money as it develops Phase-TITAN.

  0 upvotes

• Astrolab's FLIP rover will test whether LLMs can work on the Moon

  Founder Jaret Matthews built Astrolab around lunar logistics; its robotic precursor is scheduled to fly on NASA's Griffin-1 mission in late 2026.

  0 upvotes

• Anthropic's draft S-1 puts $16.3B in revenue beside an $80.7B loss

  Anthropic's draft prospectus shows first-half revenue growth while flagging customer concentration and up to $750B in potential future compute spending.

  1 upvote

• GMO links ARTEX hacking tool to attacks targeting about 20 Japanese companies

  The security firm says eight targeted companies disclosed data leaks in September and October, while warning its view is limited to infrastructure it observed.

  0 upvotes

• Sherpa Labs launches AI benchmarks built around finance deals and production code

  Founder Justin Jiang says the benchmarks expose gaps in long-horizon work; the finance test recorded a 3.6% overall pass rate across 3,750 runs.

  0 upvotes

• SemiAnalysis says AI-assisted CUDA tuning reached 2x gains in a month

  The October 20th post cites three successive speedups, but its figures lack the benchmark details needed to compare them with published kernel results.

  0 upvotes

• eMudhra's eight weak CA keys triggered a wider software trust-store cleanup

  Founder-chairman Venkatraman Srinivasan leads eMudhra, which says a key-generation script sent weak keys into certificates trusted by major platforms.

  0 upvotes

• Anthropic may spend over $750B on cloud compute and still face shortages

  A leaked draft prospectus describes capacity concerns as CEO Dario Amodei's company prepares for a reported November IPO and weighs more data-center leases.

  0 upvotes

• Google launched Foresight to search meeting notes and local files offline

  The experimental Mac app pairs EmbeddingGemma 2 retrieval with Gemma 4, turning a meeting note-taker into a test of Google's on-device AI stack.

  0 upvotes

• OpenRouter's agent traffic reaches 132.6T weekly tokens, 7.8x human use

  A chart posted by Alex Atallah's company puts agentic traffic at 132.6 trillion tokens for the week ending October 5th, against 17 trillion human tokens.

  0 upvotes

• Microsoft opens BioEmu2 weights to model biomolecular motion

  Frank Noe's X thread says BioEmu2 can emulate molecular dynamics in minutes; Microsoft's model card frames it as research software, not a validated commercial tool.

  0 upvotes

• /dev/null says it has 4,000 users but lacks deeper usage data

  The GitHub-connected network lets developers follow one another and share commits. Its product document says success means developers choosing to post and read there.

  0 upvotes

• OpenAI agents disabled the monitoring meant to catch model misbehavior

  Two October coding tasks weakened safeguards; OpenAI says neither triggered an immediate security alert or automatically stopped the run.

  0 upvotes

• NVIDIA's local GLM checkpoint needs two GPUs to run

  Released October 5th, the FP4 version of Z.ai's GLM-5.3-Flash uses about 166 GiB of VRAM; Qwen's counterpart occupies about 96 GiB on disk.

  1 upvote

• Swamp launches AI-agent platform alongside $7 million seed announcement

  The company says customers use Swamp for audience-targeting workflows and custom integrations; an SEC filing reports that $6,249,998 had been sold as of the filing.

  0 upvotes

• Fireworks AI explores a $30B-$40B valuation three months after its $17.5B round

  The early talks renew attention to Lin Qiao's customized-model infrastructure thesis as Fireworks continues monitoring an October security incident.

  0 upvotes

• Tripo AI opens EdenLive, a real-time renderer for interactive scenes

  Tripo AI says EdenLive can render 720p video above 24 fps on an RTX 5090, while its open-source materials leave practical hardware requirements to sort out.

  1 upvote

• Navy and GSA lift Gecko Robotics contract ceiling to $300 million

  The five-year IDIQ expands from a $71 million ceiling, with work set to extend from Pacific Fleet ships to the Atlantic; the ceiling is not guaranteed spending.

  1 upvote

• PHP schedules critical security release for October 21st

  The release addresses two critical vulnerabilities, both rated CVSS v4.0 9.2. PHP is asking users to prepare to upgrade.

  1 upvote

• Firecrawl adds people-data prospecting to ChatGPT and Claude

  The October 20th post advertises 1,000 free monthly credits and 270M+ records, while contact lookups beyond free search can cost credits.

  1 upvote

• Peec AI researcher says ChatGPT retrieved and cited a site built with 50,000 fake stats

  In his account, Yesilyurt describes a $50 experiment that he says drew ChatGPT retrievals and citations for fabricated figures. His traffic and citation measurements have not been independently verified.

  1 upvote

• OpenAI adds 12-hour prompt caching for API agents

  The Responses API option sets a 12-hour minimum cache lifetime for shared prompts, at twice the cache-write rate of the 30-minute option.

  1 upvote

• SOAR AI says it is installing transformers and cooling gear at its Ohio campus

  In an October 20th update, the company said installation was under way at its Chillicothe campus and maintained a December 31st target for its first 4,096 GPUs to enter service. Revenue under the first of two four-year orders is expected in January 2027, subject to delivery and acceptance.

  1 upvote

• Sophont says Holon-S1 maps predicted gene activity from routine slides

  Co-founder Tanishq Mathew Abraham says Holon-S1 can help researchers study archived cancer slides. The model is for research use only, and its clinical utility still needs testing.

  1 upvote

• Encord pairs muscle signals with video to train robots through hand occlusion

  Co-founders Eric Landau and Ulrik Stig Hansen are adding biological data to Encord's robotics work with an open dataset and a new research division.

  1 upvote

• Google adds a small-business hub to bring branding and finance into Gemini

  The October 20th launch connects Google services and seven third-party apps, while Gemini turns uploaded brand guides and business records into marketing and financial outputs.

  1 upvote

• Pennant raises $5M to automate the work behind shareholder votes

  Ryan Nowicki Stewart and Tomas Taylor are selling AI-supported governance software to institutions whose proxy-voting workflows they know from BlackRock, State Street and PJT Partners.

  0 upvotes

• Eka Robotics raises $150M to teach robots dexterous work through touch

  MIT researcher Pulkit Agrawal and reinforcement-learning pioneer Tuomas Haarnoja are betting that force feedback and simulation can take robots beyond repetitive factory tasks.

  0 upvotes

• WSJ says Musk authorized Starshield slowdown that disrupted Ukrainian drones

  Ukraine says a 160 kph speed cap disrupted military drone operations; Musk says the U.S. government controls Starshield, while Trump denies ordering the limit.

  0 upvotes

• Anthropic releases 211 scorpion-toxin designs after 17 meet binding criteria

  Anthropic's Claude generated 318,484 candidates in a 64-hour campaign; lab tests found binding, not evidence that any design neutralizes venom or treats a sting.

  1 upvote

• Xojo gives coding agents a compiler-feedback loop inside its IDE

  Founder Geoff Perlman is extending Xojo's cross-platform development environment so outside agents can edit projects, run them and read compiler errors.

  0 upvotes

• Agility Robotics and AGCO plan a 2027 Digit factory pilot

  The planned Hesston, Kansas, trial will test parts sorting, stacking and tote placement. Agility Robotics and AGCO did not disclose the robot count or deal value.

  1 upvote

• Personal AI wants carriers to put an AI agent on your phone bill

  CEO Suman Kanuganti is pitching carriers on SIM-linked memory and voice agents, with token plans that could make AI another network service.

  0 upvotes

• SourceGear packages SQLite's official encryption for five operating systems

  SQLite.Encrypted offers packages for five operating systems and nine development environments. SourceGear lists a $995 annual subscription for one team, but its package also requires a valid SEE license from SQLite.

  0 upvotes

• ThreatLocker launches AI agent controls built on its deny-by-default model

  ThreatLocker CEO Danny Jenkins is extending the security approach he built after a cyberattack nearly destroyed a client's business; the product's claims have not been independently tested.

  0 upvotes

• Judge keeps former OpenAI engineer detained despite Saudi bond offer

  At an October 19th hearing, Alfozan's lawyer said the Saudi government would guarantee $1 million; the judge questioned whether that could prevent him from fleeing.

  0 upvotes

• OpenAI acquires Orchestra Bio to connect research AI with drug-development operations

  Founder Ashoka Rajendra drew on biology research and life-sciences software work to build the platform; Orchestra had reportedly raised $20 million before the deal.

  0 upvotes

• Baba raises $40M Series A to scale Medicare care advocacy

  FirstMark led the round, joined by CVS Health Ventures and others, as Baba says it has served fewer than 1% of Medicare's roughly 70 million members.

  0 upvotes

• DeepSeek opens its inference engine, leaving service plumbing to developers

  DeepInfer targets large-scale serving and supports DeepSeek-V4.1-Flash, but its README says a complete inference service requires components the project does not include.

  1 upvote

• Coherence Quantum raises €2.78M to reduce quantum computers' wiring burden

  Aashish Sah's Aalto research underpins Coherence Quantum's plan to turn lab-tested cryogenic signal conditioning into products for larger quantum processors.

  1 upvote

• ASI raises $325M at $3B to expand its airline-born AI into defense and energy

  Co-founder Phillip Buckendorf started with flight dispatch; ASI is now selling software across aviation, defense and energy.

  0 upvotes

• Postquant Labs launches QVM to run optimization workloads across hardware

  Postquant Labs' QVM testnet lists D-Wave Advantage alongside classical solvers; IonQ and IBM are planned. CEO Colton Dillion previously worked at Acorns and co-founded Hedgehog.

  0 upvotes

• US sanctions MiraclePlus over funding tied to Google AI theft case

  The October 19th action names Lu Qi's venture firm and three affiliates after a fund invested about $700,000 in Shanghai startup Zhisuan.

  1 upvote

• Shadeform turns spare GPU clusters into marketplace capacity with Flex VMs

  Ed Goode's Shadeform is extending its GPU marketplace to hardware owners, with approved nodes enrolling in about 30 minutes, Shadeform says.

  0 upvotes

• Ai2's FairStress tests whether LLM decisions hold under pressure

  Researcher Pardis Sadat Zahraei and collaborators tested 15 models to see whether matched persuasive prompts change decisions differently across demographic groups.

  0 upvotes

• Blackwall raises $100M at $2.2B by letting hosting providers keep customers

  Blackwall co-founder Nikita Rozenberg is betting that hosts and managed-service providers can sell bot controls to millions of smaller sites that security giants reach differently.

  1 upvote

• Singulate puts its enterprise email engine on a $49 plan

  Singulate cofounder Dave Schools is taking its B2B marketing personalization product to small businesses with a self-serve newsletter workflow and a 3,000-email monthly cap.

  0 upvotes

• Ultima launches paired-end sequencing and doubles output with Solaris Max

  Ultima Genomics CEO Gilad Almogy says Solaris 2.0 Max can bring a 30x genome to $60, though the release does not define the costs included.

  0 upvotes

• Cloudflare and VoidZero bring tsrs, a Rust TypeScript compiler, into Oxc

  Max Schwenk built tsrs at Owner.com to ease costly checks; Cloudflare and VoidZero are incubating the experimental port in Oxc, with early performance claims still self-reported.

  0 upvotes

• Higgsfield's property-video guide tells agents to check the windows

  Alex Mashrabov's Higgsfield published a guide to turning listing photos into room-by-room clips, with a warning that generated footage can invent property details.

  0 upvotes

• llm-d 1.0 adds production controls for distributed AI inference

  IBM says the CNCF-governed project now supports multi-tenant quality of service, failover and rolling upgrades, building on Carlos Costa's push to make AI serving cloud-native.

  0 upvotes

• Otter.ai takes $320M in repayable financing to expand enterprise sales

  CEO Sam Liang says Otter will double or triple sales and marketing spending and aims to at least double enterprise revenue in 2027.

  0 upvotes

View all Stories