The AI Wire ●

A security researcher, Sunil Tripathy, published a disclosure on October 15 describing a flaw in ChatGPT's GitHub integration: hidden instructions in a public repository allegedly led the agent, during a routine review request, to modify a different private repository and trigger a GitHub Actions workflow.

In the researcher's test, which used accounts he controlled and a dummy secret, the workflow sent the test value to an external server. He assigns the finding a CVSS 4.0 score of 9.3; that rating is his own assessment, not a vendor classification.

Treat this as an unverified report for now. The write-ups all trace back to the single disclosure, the findings have not been independently verified, and the coverage includes no confirmation from OpenAI. The pattern it describes is indirect prompt injection: hostile instructions arriving inside material the assistant is asked to read.

A new model identifier, azure_ai/claude-fable-5-5, appeared in LiteLLM's model data on October 13, sourced from an Azure Models API listing for the East US 2 region. The listing marks the model as generally available with a December 2027 deprecation date, but nearly every other field is inherited from Fable 5.1, and no source yet states a price, capability or context figure specific to Fable 5.5.

A tipster cited by Wccftech said Fable 5.5 is slated to land this week, possibly as soon as today, barring late hold-ups. Separate chatter claimed a quiet grey-scale test in the Claude web interface, but those reports are anecdotal and unverified.

Treat this as rumor-grade for now. Anthropic's own documentation, pricing pages and newsroom still list Fable 5.1 as the newest Fable model, with Haiku 5.5 on October 7 as its most recent model announcement, and independent trackers have no Fable 5.5 page. An infrastructure listing is a classic pre-launch tell, not a launch.

Update October 16: a leaker account, Lyra, claimed on October 15 that partner companies have received Fable 5.5 for testing and guessed at a launch next week, while stressing there was no release date. Separately, the LiteLLM listing that started this rumor had been removed by October 16. Anthropic has still announced nothing, and Fable 5.1 remains its newest Fable model, so this stays unresolved.

Update October 17: TestingCatalog reported that Fable 5.5 has reportedly reached Anthropic's partners, with six early-access codenames spotted in the Devin CLI, plus one in a GitHub model picker, suggesting it may not be the only model on the way; the codename Onigiri is tipped as the favourite for Fable 5.5. Anthropic has confirmed nothing, and a launch next week is presented as plausible, not scheduled, so this stays unresolved.

The Information reported on October 12 that OpenAI is in talks to revive Sora, the video model it shut down earlier this year — the consumer app, website and API were closed in March and a reported $1 billion Disney deal died with them — through a stealth startup co-founded by Hollywood producer Jeffrey Katzenberg, former Sora research lead Bill Peebles, and former Dropbox CFO Sujay Jaswa.

Under the talks, the startup would get deep access to customize Sora and would sell software or services built on it to the film industry, aimed at professional filmmakers rather than the general public. OpenAI would take an equity stake if a deal closes. It is unclear whether the startup would receive Sora's model weights, though sources say the access discussed would be deep enough for extensive customization.

Nothing has been signed and there is no timeline. This is a single-source report about private negotiations, with no comment or announcement from OpenAI, so it stays rumor-tier until the companies say something themselves.

Daytona CEO Ivan Burazin shared an estimate, citing Kenneth Auchenberg's math, that Meta's Muse agent needs roughly 65,000 CPUs and 75 petabytes of DRAM — about $2.8 billion in infrastructure — to serve 100 million users. Scaled to Facebook's 3 billion-user footprint, the back-of-envelope figure reaches some $84 billion, more than Meta's entire 2025 capex.

The estimate landed hard in tech-exec circles: Box CEO Aaron Levie picked it up with the warning that "the compute needed for the stage of AI we're about to enter is going to be insane," and Elon Musk replied "People have no idea." It tracks with what is known — each Muse user gets a cloud VM with 8 GB of RAM for their agent — and with a separate report that Muse drove renewed server-CPU demand on AMD EPYC hardware after its September 8 launch.

Caveat: this is one executive's envelope math amplified by X, not a company disclosure. But it is the first concrete unit-economics number attached to Muse's free tier, and it reframes the agent race as an infrastructure-spend race — exactly the lens the labs themselves are starting to use.

Nvidia is in early talks to deepen its investment in Reflection AI or acquire the open-weight startup outright, the Financial Times reported Saturday, citing people with direct knowledge of the matter. Reuters carried the FT report the same day, and Bloomberg also confirmed the talks.

Talks are at an early stage and could take several forms, including an acqui-hire in which Nvidia hires Reflection's staff and licenses its technology rather than buying the company outright. That structure could avoid a lengthy regulatory review; Nvidia used the same approach late last year in its $20 billion Groq deal.

Reflection was founded in 2024 by former Google DeepMind researchers Misha Laskin and Ioannis Antonoglou to build frontier-grade open-weight models in the US. It released its first model, Beam, on October 5, and Nvidia already holds an $800 million stake; the startup was valued at $25 billion in March.

Reuters could not independently verify the report, and both companies declined or did not respond to requests for comment. The discussions could still fall apart, but an agreement could come within weeks.

TechCrunch reported Monday that inference-chip startup Etched is fielding early investment offers valuing the company at $40 billion to $50 billion, according to people familiar with the bids. That would be roughly double the $21 billion mark set in September, when Jane Street led a $700 million round, itself only two months after a $300 million Sequoia round at $10.3 billion.

The pitch underneath the numbers is a claimed $1 billion customer order book and a first delivered system to Jane Street, which is both an anchor investor and an early customer. Etched says its chips process tokens faster and cheaper than Nvidia's, and it is trying to prove that at scale from a 10-megawatt Silicon Valley data center and a facility near TSMC in Taiwan.

The talks are early and terms could change, with TechCrunch noting it is unclear whether the offers are primary or secondary, and Etched declined to comment. Treat as rumor until a financing is confirmed.

An anonymous model listed as stealth/space-bunny-alpha on OpenRouter since September 23 has quietly become the most-used model on the platform, logging a reported 5.2 trillion tokens on October 1 — more than GPT-6 Astra on its busiest day. It is also available via OpenCode Zen as space-bunny-free and through BeatAPI. No company has claimed it.

Independent testing describes a capable model: 1M-token context, up to 524K output tokens, adjustable reasoning effort, tool calling, and image and video input. Third-party evals found it shipping working Three.js scenes, finding needles in 91K-token documents, and holding its own on coding tasks — though it reportedly never checks its own output.

The strongest clue yet points at MiniMax. Two independent tokenizer studies — one matching all 50 test strings to eight tested MiniMax models — put Space Bunny in the MiniMax family, and MiniMax quietly shipped an M3.1-Flash-Preview inside its MiniMax Code product on September 27 without connecting it to the bunny. But the developer remains officially anonymous, the free preview has no published end date, and OpenRouter notes the provider may retain prompts. Fingerprint similarity is not confirmation — the rumor stays unresolved.

Update, October 5: the preview ended. SpaceBunny's FAQ records that the OpenRouter route (stealth/space-bunny-alpha, listed September 23) was withdrawn on October 5, 2026 — an October 6 check found no available endpoints and returned 404. The maker was never officially confirmed; the strongest fingerprint evidence still points at MiniMax's M3.1-Flash-Preview, but MiniMax never confirmed the link. The underlying claim — that an anonymous free model quietly topped OpenRouter's usage charts — was verified by multiple trackers while it lasted.

OpenAI appears to be preparing tiered developer plans spotted in an onboarding flow, according to TestingCatalog via RuntimeWire: Free, Prototype, and Accelerate. The Free tier would offer limited API access through Codex and the Playground, while Prototype adds all models with higher usage.

The Accelerate tier is quoted at "Starting at $50" and targets production apps, with access to all models and higher usage and rate limits. The report stresses this is an onboarding screen, not a launch announcement — the billing period and what's bundled are unresolved.

If it lands, this would be OpenAI's first packaged developer ladder, moving new builders from experimentation to production without navigating metered API setup first. Sam Altman has been hyping a developer push ahead of DevDay on Tuesday, telling developers to "see you next week."

DEBUNKED 2026-10-01: OpenAI DevDay on September 29 came and went without any Free, Prototype or Accelerate developer plans. Post-keynote recaps of the pre-DevDay leaks (Bitsminds) explicitly list the tiered developer plans among the rumors that did not materialize, and OpenAI has announced no rollout date since. The onboarding-flow evidence was real, but the predicted DevDay launch did not happen.

Update (2026-10-08): Still unresolved: the Free/Prototype/Accelerate developer plans did not show up at DevDay 2026 (Oct 6). OpenAI did introduce new pricing moves instead (Pro 500 at $500, Ultrafast tiers, Sign in with ChatGPT for partner tools).

X user @blankspeaker posted on September 26 what they described as a "very early look on Android" of music generation inside Grok Imagine. SpaceXAI's Imagine tool currently covers image and video generation with audio; its published materials do not describe a standalone music generator.

The poster says the system produced different kinds of tracks from prompts, but admits adding "a thing or 2" to the result, so the clip is not a clean demo of raw output. The post names no model and does not establish whether the audio was produced separately from a video. Treat it as an early test hint, not a product announcement.

Tibo Sottiaux posted September 26 that OpenAI is resetting usage limits for all paid users across Codex and ChatGPT Work, writing “o yes… we're back in action and we'll reset usage limits for all paid users across codex and ChatGPT work — sorry about the brief disruption!” The post apologizes for an outage that appears to have cut into paid usage earlier in the day.

Sottiaux has become a reliable public signal for OpenAI's launch and outage cadence — his posts telegraphed the ChatGPT reset timing and the GPT-6 Sol rollout earlier this month. This is single-source social chatter rather than an official OpenAI announcement, but it reads as a promise being kept: limits are being reset for paying customers after the disruption.

For paid subscribers on Pro and above, the practical takeaway is that any usage burned during the outage window should be credited back. Watch for whether the reset lands uniformly across tiers or favors particular products in the coming days.

Resolved September 30: confirmed. OpenAI reset every Codex user's usage allowance worldwide at DevDay on September 29, per the company's announcements — matching the substance of Sottiaux's Sept 26 post, which explicitly promised resets across Codex and ChatGPT Work after the disruption.

Google is reportedly testing a new 'Gemini Tasks' mode that could handle work across desktop apps, with Computer Use permissions and Gemini Live built in. The finding comes from Android Authority, surfaced in news roundups on September 25.

A desktop Tasks mode would put Gemini in direct contention with the agentic computer-use tools Anthropic, OpenAI, and Meta are all pushing this year. Until Google acknowledges the testing, it stays rumor.

Resolution (October 13): confirmed in shipped form. Google launched a tasks mode in the Gemini app for AI Pro and Ultra subscribers alongside Gemini 4 Argon, letting Gemini keep working while the user is away across more than 40 connected apps. The shipped version is an in-app toggle rather than the desktop Computer Use mode described in September reports, so the feature landed, in a different form than first rumored.

On September 25, a long-term ChatGPT Pro 20x subscriber posted on OpenAI's own community forum that hard usage limits had appeared on the plan without advance notice. The subscriber said the Pro model option became unavailable, with ChatGPT showing a "Limit reached. Try again after 4:59 PM tomorrow" message.

The post says the documentation now refers to hard daily and weekly caps, with no visible usage meter before the limit hit and no email or in-product notice to existing subscribers. At launch, ChatGPT Pro was sold at $200 a month with "unlimited usage," the poster notes, and describes the caps as a product regression rather than a plan change. OpenAI has not publicly responded to the report.

UPDATE Sept 29: Confirmed in substance. OpenAI product lead Tibo Sottiaux announced the reopening of $200 Pro sign-ups with an official new usage calculation — tacit confirmation that usage math on the 20x plan has changed. Open questions (30-day notice, grandfathering) remain unanswered, and the original subscriber report was never contradicted.

Fortune reported Thursday, citing multiple sources familiar with the plans, that OpenAI is set to preview GPT-6 Cyber, a cybersecurity-focused model, within days — with its DevDay conference in San Francisco on Tuesday, September 29, as the likely stage. Reuters carried the report on September 24. A limited group of customers in the application-only Daybreak Red program already has access for alpha testing.

GPT-6 Cyber would be OpenAI's fourth security model of the year, following GPT-5.4 Cyber in April, GPT-5.5 Cyber in June, and GPT-5.6 Cyber in August. Alongside it, OpenAI plans a first-of-its-kind enterprise security platform — a gateway environment for building automated defenses and patching vulnerabilities, modeled on how ChatGPT serves as the consumer gateway to frontier models.

Coverage has intensified ahead of DevDay: TechGig and Forkast now report the Sept 29 preview as the working plan, with chief revenue officer Dali Rajic leading a commercial push behind the security line. One caution from The Hack Academy: OpenAI's official DevDay page confirms the event and a livestreamed keynote but says nothing about GPT-6 Cyber — so the preview date remains unconfirmed by OpenAI itself.

Resolved September 29: confirmed in substance. No 'GPT-6 Cyber' model was previewed, but the enterprise-security side landed hard: Codex Security Cloud (scheduled repo scanning with dedup and cloud-prepared fixes, Daybreak defensive models included) plus Private Intelligence — zero data retention with Private Safety Processing — and a Private Inference preview coming this fall.

Update (2026-10-08): Still unresolved: GPT-6 Cyber did not appear at DevDay 2026 (Oct 6). Fortune's later reporting said it is expected "within weeks, not days." DevDay's security announcement was Codex Security Cloud instead.

Update: DevDay 2026 (September 29) came and went with no GPT-6 Cyber preview and no enterprise security platform. OpenAI's developer event centered on Dots, Codex updates, the Pro 500 tier and the Decisions API. Marking this rumor debunked.

The Information reports, citing unnamed sources, that Anthropic, Google and OpenAI are working to launch a standards body for frontier AI by the end of this year or early next — one in which the model builders would regulate themselves, without government oversight. The group would support third-party pre-deployment testing, set rules for reporting safety and security incidents, codify voluntary safety commitments, and define qualifications for independent auditors.

The labs reportedly first aimed for a public-private partnership, but that effort stalled under the Trump administration, per the report. The same three labs plus Microsoft founded the Frontier Model Forum in 2023 with many of the same goals; that nonprofit remains active and is said to be in contact with the new working group. Critics worry the new body could be used to box out open-source developers and smaller competitors.

In parallel, Bloomberg reports Anthropic CEO Dario Amodei and OpenAI CEO Sam Altman addressed a UN Security Council meeting on AI this week and agreed on the need for global standards around AI risk. The UN appearance is real; the standards-body plan itself remains single-source and unconfirmed.

Resolved September 30: confirmed in substance. On September 29, Trump hosted the AI chiefs — Amodei, Zuckerberg, Pichai, Altman — at a White House lunch and they signed the 'White House Accord on Super Intelligence': four layers of self-regulation described as 'morally binding.' That's the labs regulating themselves via a shared standards framework, exactly what the rumor described.

New identifiers — openai/gpt-6-sol-pro and openai/gpt-6-luna-pro — appeared in OpenRouter’s catalog within the last day, alongside the base Sol and Luna slugs that launched September 22. pricepertoken.com’s New Models Today tracker lists all four, each timestamped “OpenRouter · 9h ago”, and OpenRouter’s own GPT Sol Latest page names “GPT-6 Luna Pro, GPT-6 Luna, GPT-6 Sol Pro” among OpenAI’s text models.

UPDATE 2026-09-23 (morning): confirmed — but as routing conventions, not new models. OpenRouter’s own GPT-6 Migration Guide now officially documents a “pro reasoning” variant for every tier of the GPT-6 family (Astra, Sol, Luna): “Pro reasoning is selected with reasoning.mode: “pro” on a standard slug or by using the -pro slug, the same as the GPT-5.6 family.” So the -pro slugs are real, supported routes to the same models with higher-effort reasoning — at GPT-5.6-era precedent, identical per-token rates, more cost per task. The leak pattern matched the pre-Sol/Luna launch, but this time it signaled routing, not a launch.