Highlights
What happened at the companies we cover, by day, newest first. Every item links to its sources.
Showing 1–20 of 38 highlights
Monday, Nov 9 38 stories
Abridge lays off staff in an 'organization-wide realignment'
NewsNewsXBluesky
Abridge confirmed on Nov 9 that it has laid off an undisclosed number of employees as part of what it called an organization-wide realignment meant to flatten the company. A spokesperson told KRON4 that affected staff get transition packages and outplacement support, and said the company remains well capitalized and still holds more than half of the capital it has raised. Pittsburgh Business Times also reported the cuts and said the impact on Abridge's Pittsburgh footprint was not disclosed. On X, health-IT watchers said health systems still mid-rollout may push for named implementation and training support in renewals.
Sources
Lovable acquires the Radish AI team to build agents for businesses
NewsXNews
Lovable said on Nov 9 that it has acquired the team behind Radish AI, a startup that built infrastructure for building, deploying and scaling complex AI agents. The three co-founders, whose backgrounds include Meta, Google, Bloomberg, JPMorgan and Goldman Sachs, join Lovable's engineering team across Stockholm and London. Radish co-founder Nirman Gupta will lead engineering on agents for businesses from the new London office, which Lovable calls its second engineering, product and design hub. CEO Anton Osika and Gupta both announced the move on X, and UK outlet BusinessCloud covered it; deal terms were not disclosed.
Sources
Engineers debate LLMJacking and model-weight theft after the Fireworks breach
DiscussionXHN
The breach update set off discussion on X about how attackers profit from AI infrastructure. One commentator compared it to crypto-style attacks, arguing the attacker monetized compromised keys by selling inference on routers. A customer shared the notification email and asked whether this is the first public incident where model weights were targeted, and Fireworks CEO Lin Qiao replied that attackers did not appear to be after weights and most downloads were partial and stopped. On Hacker News, one commenter called the breach potentially company-ending.
Sources
1 more source
Fireworks says Oct 24 intrusion exposed billing data, some datasets and partial model weights
NewsNewsHNXBluesky
Fireworks updated its October security bulletin on Nov 9, saying a third-party cloud provider hosting its R&D machines was breached on Oct 23 and the attacker used that access on Oct 24 to reach Fireworks development machines. The company now says account and billing information was affected, and for a limited set of accounts also secrets, datasets and incomplete model weights, and it links the attack to a financially motivated LLMJacking group. Fireworks says it found no impact on Inference, Training SDK or Nexus, which are zero data retention by default, and it has disabled self-serve dataset uploads and downloads for Batch Inference and dataset-based training while it investigates. Bloomberg reported this is the company's second known hack in weeks and quoted a spokesperson saying it is working with customers including Notion and Cursor. A Forbes reporter posted that one customer described poor communication and said it was moving off the platform.
Sources
1 more source
Cursor adds Claude Fable 5.2, which tops CursorBench at max effort
LaunchXOther
Cursor made Anthropic's Claude Fable 5.2 available on Nov 9, and its CursorBench 4.0 page now ranks Fable 5.2 Max first at 60.0%, with the High setting at 56.9% versus 49.2% for Fable 5.1 High. Cursor's model docs list pricing of $8 input and $40 output per million tokens, down from $10 and $50 for Fable 5.1, and advise starting at high effort instead of reusing old Fable 5.1 settings. The model is off by default for Privacy Mode teams and all Enterprise customers, who need an admin to opt in to Anthropic's data-retention terms, and requests that trip its guardrails fall back to Claude Opus automatically. Cursor's launch post was its most-engaged post of the day on X, and users posted early impressions on efficiency and usage limits.
Sources
Modal co-authors study finding many small agent teams beat best-of-N and one big team
DiscussionX
Researchers from Duke, Modal Labs, Princeton and NYU published a paper arguing that test-time compute for coding agents scales best by sampling many small teams that share a forum and keeping the best team's result. With a 27B Qwen model and a budget of 100 agents on 12 long tasks, ten teams of ten beat 100 independent agents on 10 of 12 tasks and beat a single 100-agent team on all four tasks where that was run. At equal dollar cost the teams beat GPT-5.6-sol on half the tasks tested, and teams with a manager agent did worse than flat teams. The agents ran in Modal sandboxes, and posts on X summarizing the results drew modest engagement.
Sources
Well-known systems engineer rejoins Modal after a stint at Thinking Machines
HiringX
A former Modal engineer announced on X that he is back at Modal after working at Thinking Machines since last year, where he describes working on ML and production systems and on kernels for training and inference. The post drew more than 1,600 likes, making it by far the most engaged Modal item of the day. Replies congratulated Modal, with one joking that the company is collecting top talent and linking the return to his recent inference experimentation. The hire adds frontier-lab systems experience to Modal's inference and infrastructure team.
Sources
Perplexity adds Claude Fable 5.2 to Computer and Max Search, citing a record WANDR score
LaunchXNewsBluesky
Perplexity made Anthropic's Claude Fable 5.2 available on Nov 9 in Perplexity Computer for Pro and Max subscribers, and in Search for Max subscribers only. The company says Fable 5.2 scored a record 0.851 on WANDR, its own wide-and-deep research benchmark, at $14.51 per task, putting it on the benchmark's cost-performance Pareto frontier. AlphaSignal notes that the main benchmark claims come from Anthropic and Perplexity and need independent replication. It also notes the announcement says nothing about API availability or whether Fable becomes the default model. Perplexity's X post drew strong engagement, and some posters flagged that the score is self-reported.
Sources
Crusoe Cloud adds Google Workspace SSO enforcement and an Okta SCIM guide
LaunchOther
Crusoe Cloud's changelog shows organizations can now enforce Google Workspace as their single sign-on provider, configured in IAM under User Authentication once the account team enables SSO. The same day Crusoe published a guide for provisioning Crusoe Cloud users from Okta via SCIM, covering app connection, mapping organization roles, and rotating SCIM API keys. These sit alongside recent Managed Kubernetes updates in the changelog. The changes were posted only in Crusoe's docs with no wider coverage.
Sources
Debate continues over what Google's takeover of Crusoe's Wyoming site means for neoclouds
DiscussionXBluesky
Social discussion of Google taking over Crusoe's southeast Wyoming data center site kept going on X today. Chamath Palihapitiya, in a clip shared on X, argued that AI labs without their own power and data centers are effectively renters, using the Crusoe site as his example. Another poster framed the deal as Google preferring to own a site with power rather than pay a developer rent on it, after Crusoe spent over a year on land, permits and a power plan. A separate post noted that Google's $17 billion figure refers to its broader Wyoming investment plan, not the undisclosed price paid for the site.
Sources
Investors keep debating Nscale's Microsoft exposure as Microsoft trims neocloud partners
DiscussionXNews
Discussion on X on Nov 9 kept circling Microsoft's decision not to take up an option to more than double its Nscale compute lease in Ward County, Texas, a deal the FT said would have been worth tens of billions. Silicon UK's write-up notes that the existing contract stays in place, that other customers such as Figure have signed for the site, and that Microsoft earlier left Nscale's Essex project over power delays. Posters linked the move to wider strain on neoclouds, including bond-market access and credit spreads, and many framed delivery timelines rather than demand as the real issue. The tone was mostly negative, with a few posters arguing that Microsoft is cutting suppliers that are late, not cutting compute.
Sources
Norway pauses new data-center grid reservations, raising doubts over Nscale's Narvik expansion
DiscussionNewsX
Norway's grid operator Statnett said on Nov 9 that it is temporarily halting new grid-capacity reservations for data centers, with the pause expected to lift by Aug 1, 2027 at the latest; existing connections are not affected. Statnett said 4,200 MW of data-center capacity already has reservations and 9,100 MW sits in the queue. Norwegian business daily DN framed it as a cold shower for Nscale, whose Nordic chief argued Norway needs more AI compute. A widely shared X analysis said Nscale's first 230 MW Narvik phase already has power reserved, but later phases look at risk.
Sources
Runway launches Sightline for 360 location panoramas and any-angle shot framing
LaunchX
Runway released Sightline on November 9, an App Studio tool that generates a 360-degree panorama of a described location and lets creators frame shots from any angle while the environment stays consistent across them. Panoramas can be exported for 3D use as scene environments. The Sightline app exposes controls such as camera height, time-of-day lighting, up to five reference images, lens focal length, aspect ratio and output resolution up to 2160p. Runway's announcement on X drew modest engagement from its creator audience, alongside a separate post inviting creators to a new Runway Nights event series in New York.
Sources
BFL's Europe API serves only a reduced model set, pushing high-res video to the global endpoint
OtherBlueskyOther
Black Forest Labs opened a status incident on Nov 9 at 11:30 UTC saying its Europe endpoint, api.eu.bfl.ai, is serving only a reduced set of models. FLUX 3 image, Klein and HD-only FLUX 3 video still work there, along with most FLUX.2 and FLUX.1 endpoints. Higher-resolution FLUX 3 video, video continuation, the video upscale and edit tools, and some FLUX 1.1 Ultra options require the global api.bfl.ai endpoint. The incident stayed in monitoring with the Europe API marked degraded. It follows an earlier Europe compute outage that BFL says it fixed on Nov 5.
Sources
Arena posts same-day Code Arena scores for Fable 5.2 and Muse Ember, adds Inkling 2 Small
OtherXOther
Arena's leaderboard changelog for Nov 9 lists four additions: Claude Fable 5.2 (xHigh) and Muse Ember (Max) on Code Arena WebDev, Inkling 2 Small on Agent Arena, and Kimi K3.1 Pro on Text Arena. The WebDev board shows Fable 5.2 at 1832 under an AutoEval tag with no votes yet, above Opus 5.5 (Max) at 1818. Muse Ember is listed as a preliminary entry at 1608 with $0.10/$0.50 pricing. Arena's own X posts calling Fable 5.2 the new WebDev #1 and Ember the top small model were its most-engaged posts of the day, and prediction-market accounts on X tracked the Text Arena standings.
Sources
1 more source
Harvey expands Wolters Kluwer deal to bring CCH iKnowConnect content into Harvey in APAC
NewsXOther
Harvey and Wolters Kluwer Australia expanded their partnership so CCH iKnowConnect subscribers in Australia, New Zealand, Singapore and Malaysia can use their licensed legal and tax content inside Harvey. The addition covers secondary material such as editorial commentary, case headnotes and practical guidance on top of the primary law already available. Subscribers can pick CCH as a knowledge source and query it alongside internal firm material, with answers linking back to the source. Harvey also noted plans for a Tokyo office in early 2027 and shared the news on its X account.
Sources
Harvey adds Review Table Sync to update document reviews as new files arrive
LaunchOther
Harvey introduced Review Table Sync, which links a Review Table to its source Vault so the table flags when new documents land from a data room, iManage or another connected source. One click adds the new files as rows and runs the existing review questions against them, while rows the team already reviewed stay untouched and answers stay cited to their source. Harvey pitches it for growing diligence data rooms, group-action claims that arrive in waves, and in-house contract portfolios. The feature is available now from a Review Table's settings menu.
Sources
Addleshaw Goddard picks Harvey for all 3,400 staff and a co-development program
NewsNewsXBluesky
London-based law firm Addleshaw Goddard will roll Harvey out to its more than 3,400 lawyers and business services staff across Europe, the Middle East and Asia starting in January. The firm's Innovation Group will work directly with Harvey's product and legal engineering teams to shape and test new capabilities, and the two plan quarterly sessions with selected clients. Harvey says the choice followed a three-month evaluation in which 400 firm staff tested competing platforms on real tasks, and that the firm's feedback already shaped improvements to Review Tables. Law360 reported the deal, and posts on X framed it as co-building rather than a plain software purchase.
Sources
Black Forest Labs signs the Paris Call for AI safety and security
NewsNews
Black Forest Labs was one of ten companies that signed the Paris Call for AI safety and security on Nov 9, alongside six governments and civil-society groups. The other company signatories included OpenAI, Anthropic, Microsoft and Mistral AI. Signers commit to a shared framework for evaluating frontier models and managing their risks, to publishing non-sensitive evaluation findings in a standard format, and to sharing information on serious incidents. The text names abliteration and the removal of safeguards as a risk, while saying open-weight development should keep its benefits. That point matters for an open-weights image model vendor like BFL.
Sources
Hugging Face signs the Paris Call for AI safety and security
NewsNewsBluesky
Hugging Face was one of ten companies that signed the Paris Call for AI safety and security on Nov 9, alongside six governments and several civil-society groups. Other company signatories included OpenAI, Anthropic, Microsoft and Mistral AI. Signers commit to a shared framework for evaluating frontier models and managing their risks, to publishing non-sensitive evaluation findings in a standard format, and to sharing information on serious incidents. The text also flags the risk of stripping safeguards from models, naming abliteration, while saying open-weight development should keep its benefits. The declaration remains open to further endorsements.
Sources