AI digest — Oct 24, 2026
Top line: Pennsylvania has signed a law requiring labels on AI-generated ads — any ad significantly modified or generated by AI, whether still image, video, audio, or text, must carry a label in the same format as the ad itself. Meta's revised business terms now reportedly require business customers to indemnify the company — including attorneys' fees — for claims tied to its AI features, adding a liability clause to a terms package that has already drawn two language removals and a November 20 enforcement date. The same day brings Block's Dessa, an AI voice model trained entirely on human speech that reportedly runs on a MacBook offline — Gizmodo has listened and calls the voices human-ish but uncanny — plus an Exiger CEO defending the "SaaS apocalypse," a brief item on the single-source Kimi K3.1 Pro explainers, and the BBC's Blue Planet III keeping its imperfect footage in on purpose, so that an over-polished picture can't be mistaken for AI.
- Meta's business terms make customers liable for its AI's mistakes
- Block introduces Dessa, an AI trained only on human speech that runs offline
- Exiger's CEO: the "SaaS apocalypse" is real, but not what people think
- Pennsylvania will require labels on AI-generated ads
- Blue Planet III keeps its rough edges in, so the footage can't be mistaken for AI
Meta's business terms make customers liable for its AI's mistakes
Sources StartupFortune
What happened: Meta's new business terms require customers to indemnify the company for claims tied to its AI features, including attorneys' fees — leaving customers holding the legal exposure for what Meta's AI does, according to StartupFortune (published about 1:00 AM ET October 24). The terms are the same package this storyline has been tracking: scraping and training language was removed in two passes (October 17 and October 22), and Meta delayed enforcement from October 30 to November 20 on October 23.
What's genuinely new: A liability-allocation clause is a new layer for this terms package. The earlier rounds turned on what content Meta could crawl and train on; this report establishes that the terms also shift legal exposure onto customers — any claims arising from the AI's mistakes, with Meta covered for attorneys' fees.
Watch: Whether the indemnity clause survives to the November 20 enforcement date the terms currently carry, how business customers respond, and whether legal experts give it the same scrutiny the scraping language drew.
Block introduces Dessa, an AI trained only on human speech that runs offline
What happened: Block, the payments company founded by Jack Dorsey, has introduced Dessa, an AI model trained entirely on human speech — without text data or pretrained models — that reportedly talks like a human and can run locally on a MacBook without an internet connection, according to an MSN report published about 4:00 AM ET October 24; the copy does not name the outlet that originally reported it, so those details stand without a named first source. Gizmodo followed at about 8:00 AM ET October 24 with the model's voice samples — many uncanny, human-ish voices — and confirms the model goes by Dessa.
What's genuinely new: Speech-only training is the claim that distinguishes the model if it holds — learning from human voice alone rather than text and pretrained components — paired with the unusual selling point of offline operation on an ordinary laptop. Gizmodo's listen-through adds the early qualitative read: the voices are convincing enough to be unsettling.
Watch: First-party confirmation from Block, details on model size and training data, latency and quality against text-trained voice systems, and what Dessa is actually aimed at.
Exiger's CEO: the "SaaS apocalypse" is real, but not what people think
Sources Yahoo Finance
What happened: Exiger CEO Brandon Daniels says the "SaaS Apocalypse" is real — but not what people think it is, in a Yahoo Finance interview published about 6:00 AM ET October 24; the conversation with Daniels took place last month, according to the article.
What's genuinely new: A sitting enterprise-software chief endorsing, with a twist, the idea that the traditional SaaS business is headed for disruption — while claiming the shape of that disruption diverges from the popular version of the story.
Watch: What Daniels' version of the SaaS shakeout actually is — the full argument and its evidence — and whether other software-company leaders echo it.
Pennsylvania will require labels on AI-generated ads
Sources The Verge
What happened: Pennsylvania Governor Josh Shapiro signed the Artificial Intelligence Deceptive Advertising Act into law on Thursday, October 22, according to a The Verge report published about 2:32 PM ET October 24. The law requires any ad — whether a still image, video, audio, or text — that "has been significantly modified or generated by artificial intelligence" to include a label in the same format as the ad itself. That applies to TV commercials from multibillion-dollar companies as well as restaurants' ugly, crowded flyers, per the report.
What's genuinely new: A state moving AI-ad disclosure out of platform policy and into statute — a legal labeling duty covering every ad format, rather than each platform's own tagging rules.
Watch: How "significantly modified or generated by artificial intelligence" is interpreted and enforced, whether other states copy the statute, and how advertisers and platforms implement a label "in the same format as the ad itself."
Blue Planet III keeps its rough edges in, so the footage can't be mistaken for AI
Sources The Verge
What happened: BBC's Blue Planet III is deliberately leaving in its less-than-perfect shots to prove its authenticity, per a The Verge report published about 5:29 PM ET October 24. Previous incarnations of the nature series were defined by gorgeous, crystal-clear footage that seemed almost too good to be real — and one dead giveaway of AI is when things look too polished and seamless. This edition goes "warts and all"; executive producer Elizabeth White has previously pointed to "the amount of plankton and things" floating in the water, per the report.
What's genuinely new: A major production treating visual polish itself as an AI tell — retaining imperfect, unglamorous footage as the proof of human origin. It is the authenticity question arriving from the opposite direction of the machine-readable marks and detectors this storyline usually tracks: instead of signing pixels, an editorial team is preserving visible imperfection so the footage speaks for itself.
Watch: Whether other productions adopt the same anti-polish stance, and whether authenticity-by-imperfection holds up as AI-generated video quality keeps improving.
Brief
- Kimi K3.1 Pro explained: A note.com post by kazu (published October 24, about 12:00 AM ET) walks through Moonshot AI's new Kimi K3.1 Pro — its performance, pricing, usage, and applications in 3D production and business. A second note.com explainer, by kuma_aihubmix (published October 24, about 3:00 AM ET), reports that Moonshot introduced K3.1 Pro in a blog post on the evening of October 23 and weighs pricing against staying on K3. Moonshot's own announcement has not been seen directly, and the explainers' pricing and usage figures are not independently confirmed, so treat the details as uncorroborated.