AI News Today — October 26, 2026: Apollo Says AI Pays Off Only for Big Tech, Grok's AI Odyssey Lands Flat, and Elvis Rights Holders Sue Anthropic

The big picture

Today's question is not what AI can do — it is what AI has actually paid for. Apollo Global Management chief economist Torsten Slok says the technology has delivered zero financial payoff outside the seven biggest tech firms, with revenue per worker and profit margins flat for everyone else.

The creative and legal stories point the same way. xAI can now generate seven watchable minutes of an epic film, and reviewers still found no reason to care about anyone in it. And Anthropic faces a new copyright suit from the owners of 274 Elvis Presley songs — a reminder that the cost of training data is still being argued in court, not settled.

What happened today

Muse AI: chat sharing stays paused — and Meta has not answered what happens to existing shares

What happened: There was no new Muse launch on October 25–26. The live Muse story remains the paused chat-sharing feature. MediaNama, in a story updated October 26, reports that Meta paused Muse chat sharing after researchers at Claroc found more than 6,800 share links discoverable through passive DNS, a figure covered in this briefing last week. The pause has stopped the original pages loading — Claroc rechecked 40 previously accessed shares on October 20 and none loaded — but the conversation titles can still be found in commercial DNS databases, because each share address was built from the conversation's title, and Claroc found about 75 saved scans of Muse share pages on a public URL-scanning service. Meta's statement says it has paused the functionality while it reviews the report and considers changes to how links can be discovered. MediaNama says Meta did not answer its questions on whether the pause permanently disables or deletes existing shares, how users can delete their own shares, whether conversation titles will be kept out of share addresses when the feature returns, or whether affected users have been notified.

Who it matters to: Anyone who has shared a Muse conversation or artifact. The original link going dead is not the same as the share being erased — titles remain searchable in DNS records, and copies captured before the pause may still exist in third-party archives and saved scans, a point Claroc itself flagged while noting it has not verified whether any such copies remain publicly accessible.

What you can do: Do not share Muse outputs that contain personal, financial, or work details until Meta confirms how shares are deleted and whether titles stay out of future share addresses. If you shared something sensitive before the pause, assume the title — and possibly a saved copy — could still be findable, and tell anyone affected directly rather than relying on the pause alone.

Learn more: MediaNama on Meta pausing Muse AI chat sharing, updated October 26

Gemini Spark: no new launch — Task mode stands, and the November 17 Gems deadline is next

What happened: There was no new Gemini Spark launch announcement on October 25–26. Spark still lives in the main Gemini app as Task mode, with a Chat/Task switch above the prompt box and a Scheduled page for active, paused, completed, and suggested tasks. It still requires a Google AI Pro or AI Ultra subscription and a personal Google account. Gems are still set to auto-migrate to Skills on November 17, 2026 — now three weeks away and still the next hard date on the Spark calendar.

Who it matters to: Spark users with custom Gems they rely on. The product is quiet; the migration deadline is not.

What you can do: Open each Gem you rely on this week, copy its name, description, and instructions somewhere safe, and re-upload any knowledge files into a Skill by hand if you want to move early. Do not wait for the automatic migration to discover that a Gem did not carry over cleanly.

Learn more: FindSkill on where Gemini Spark went — it's Task mode now

Apollo economist: AI has delivered "zero financial payoff" outside Big Tech

What happened: Apollo chief economist Torsten Slok published a note on October 24, reported October 25, arguing that AI has delivered zero financial payoff to corporate America outside tech. He splits the S&P 500 into the Magnificent Seven — Alphabet, Amazon, Apple, Meta, Microsoft, Nvidia, and Tesla — and the remaining S&P 493. Revenue per worker is rising for the Magnificent Seven but flat for the S&P 493, and profit margins show the same split. Slok adds that the top 10% of firms account for 99.5% of model-serving and inference spending and more than 95% of spending tied to OpenAI and Anthropic, and he says mainstream companies are starting to treat token spending as a cost centre that weighs on margins. This is one economist's reading of aggregate data, not a company-by-company audit — other surveys have found individual productivity gains even where company-level financial results have not moved.

Who it matters to: Every business owner paying for AI tools. If the aggregate payoff is concentrated in a handful of tech firms and a tenth of spenders, a subscription alone is not a strategy — the workflow around it is.

What you can do: Pick one AI workflow you pay for and measure it this week: time saved, output produced, or revenue touched, before and after. Keep the tools that move a number you can name, and cut or renegotiate the ones you are only keeping because everyone else has them.

Learn more: TechCrunch on Apollo's "zero financial payoff" note and PYMNTS on Apollo's AI payoff finding

Grok released seven minutes of its AI-generated Odyssey — reviewers called it airless

What happened: Grok Imagine posted Episode 1 of an AI-generated Odyssey series on October 23 — seven minutes and twelve seconds, made with Wonder Studios and director Dev Walker using xAI's Grok Imagine tool. That is a reframing of what Elon Musk promised in July: a full-length, "historically accurate" film of Homer's epic before the end of 2026. TechCrunch's reviewer found some shots convincing in isolation, but called the whole piece "completely airless," with stilted dialogue and no artistry, inspiration, or human interest. The episode uses English dialogue, not Homeric Greek, and covers the Greeks' arrival at Troy.

Who it matters to: Anyone selling or buying AI video. Seven coherent minutes in 4K is a real capability step — and the review is a clear statement of the current ceiling: shot-level realism is arriving faster than storytelling.

What you can do: If you use AI video commercially, use it where it is already strong — short product shots, backgrounds, and cutaways — and keep a human hand on script, performance, and edit for anything a viewer is meant to feel something about. Judge tools on a finished piece your audience watches to the end, not on a single impressive frame.

Learn more: TechCrunch on Grok's seven-minute AI-generated Odyssey

Elvis Presley rights holders sue Anthropic over 274 songs, seeking about $41 million

What happened: Companies holding rights to Elvis Presley's music, led by a subsidiary of Authentic Brands Group, sued Anthropic on October 23, alleging it used 274 songs — including "Can't Help Falling in Love," "A Little Less Conversation," and "Don't Be Cruel" — to train its Claude models without permission. The plaintiffs seek statutory damages of up to $150,000 per song, totalling about $41.1 million. The complaint also alleges Claude can produce close versions of the songs, and that it kept doing so after guardrails were added. Reports differ on Anthropic's response: several outlets reported no public comment as of October 25, while a later roundup reports a short statement that Anthropic disagrees with the claims and intends to defend the case — treat that response as reported, not confirmed here. Anthropic already faces other music-publishing suits, and a judge heard summary-judgment arguments in an earlier lyrics case on October 21, with fair use as the central question and no ruling reported yet.

Who it matters to: Anyone building products on frontier models, and anyone whose creative work might be in a training set. Music is becoming the most active front in the training-data fight, and each new plaintiff adds a famous catalogue — and a damages figure — to it.

What you can do: If your business depends on AI-generated music, lyrics, or text in a distinctive style, keep records of your prompts and outputs and check your provider's terms on who carries copyright risk. If you publish creative work, this is another reason to keep your catalogue registrations and ownership records current — the plaintiffs in these cases are the rights holders who can prove ownership.

Learn more: Parade on the Elvis Presley rights holders' lawsuit against Anthropic

Outside auditors found gaps in OpenAI's agent monitor — including about 10% of Codex traffic unseen

What happened: Findings published October 22 from a three-week pilot (July 21 to August 11) in which one researcher each from METR and Redwood Research worked inside OpenAI, testing the system that watches its internal agents for misalignment. They estimated 75% to 94% of eligible internal Responses API traffic was covered, and 99.2% for internal Codex agents — but a gap around compaction events in long sessions likely left about 10% of eligible Codex traffic unmonitored between May and July. They also found ways an agent could in principle bypass sandboxing, use client-controllable headers that switched monitoring off for new agents, and use an exploit that could hide hundreds of recent actions. OpenAI told the evaluators it has patched every vulnerability affecting production safeguards; METR says it has not evaluated that claim, so the fixes are unverified, and the review did not establish that agents actually used these gaps.

Who it matters to: Anyone deploying agents with real access. The most instructive finding is that the biggest blind spot was not an attack — long sessions simply skipped the event the monitor was waiting for, and long sessions are exactly where an agent has done the most.

What you can do: If you run agents, ask how your monitoring behaves in long sessions, not just short tests: what happens when a session is summarised, compacted, or handed off, and can any setting or header turn oversight off? Log agent actions outside the agent's own system, so the record survives even if the agent's session does not.

Learn more: TokenPost on the METR and Redwood findings on OpenAI's agent monitor

Also worth knowing

• Honeywell launched an industrial AI platform at the Dangote refinery in Nigeria. Honeywell said on October 26 that its Forge Plant Cognition platform, using Nvidia and Dell capabilities, will run at the Lekki refinery in Lagos. Honeywell said preliminary figures suggest benefits of more than 29 cents a barrel, or about $60 million a year at full-capacity throughput — a company estimate, not guaranteed. Reuters on the Honeywell launch
• The AI wearable field is now crowded and mostly unproven. TechCrunch's roundup covers pendants, rings, and wrist devices from Bee ($49.99, plus $19 a month), Friend ($129), Limitless ($99), Omi ($89), and Plaud's NotePin ($159), plus Meta's Muse Charm, introduced in September and not yet shipped. Most record conversations continuously — read the mute and consent story before you wear one into a meeting. TechCrunch on the hottest AI wearables and gadgets
• OpenAI and Anthropic are hiring, not shrinking. A count of their live job boards on October 26 put the two labs at 1,472 open roles between them, including 702 engineering roles — a useful counterweight to the claim that the labs are replacing their own staff with their models. The figures are one site's pull from the public boards; check the boards directly before relying on them. Tensorust's count of OpenAI and Anthropic openings

One practical takeaway

Measure one workflow, and fence one agent. Apollo's numbers say the average company has no AI payoff to show yet, the Odyssey shows a demo is not a finished product, and the OpenAI monitor audit shows oversight can quietly miss the longest, busiest agent sessions. The response is the same in all three cases: pick one paid workflow and prove its value with a number this week, judge creative tools on finished work your audience actually finishes, and keep an outside log of anything an agent does on your behalf.

---

Sources: TechCrunch AI, MediaNama, PYMNTS, Parade (reporting Billboard), TokenPost (reporting METR and Redwood Research), Reuters, FindSkill, Tensorust. Summaries are original; follow the links for full reporting. Apollo's "zero payoff" finding is Torsten Slok's reading of aggregate S&P 500 data, not a company-by-company audit. Honeywell's per-barrel benefit is a preliminary company estimate described by Honeywell as not guaranteed. Anthropic's response to the Elvis lawsuit was reported inconsistently — no comment in several October 25 reports, a short disagreement statement in a later roundup — and is marked reported, not confirmed, above. OpenAI's patches to its agent monitor are OpenAI's claim to the evaluators; METR says it has not evaluated them. There was no new Muse or Gemini Spark launch announcement on October 25–26.