The AI Wire ●

SentinelOne announced on October 29 the upcoming addition of Anthropic's Claude Mythos 5.1 to its Wayfinder Frontier AI Services offering. The service pairs security experts with frontier models from Anthropic, OpenAI and other labs to find which vulnerabilities exist in a customer's code and operations, work out which are truly exploitable today, and deliver prioritised remediation guidance.

SentinelOne calls Mythos 5.1 Anthropic's most capable model class for cybersecurity. With it, the company says, its analysts will be able to apply that model class to discover potential risks and then validate which findings are exploitable, so security teams get a clear order of operations instead of an unfiltered backlog.

The announcement lands in what SentinelOne calls "Black October" — a sharp increase in the volume and severity of vulnerability disclosures and attacks in recent weeks, including PHP2SHELL, LiteLLM and eMudhra, which its Chief Customer Officer Steve Stone cited in the release.

Day 25 of Tibo Sottiaux's 28-day Codex challenge brought a limits change instead of a reset. Tibo posted on October 29 that OpenAI is doubling all usage limits for all Pro 100, Pro 200 and Pro 500 tiers and for Business Premium, effective today and landing by midnight PST — "Graduating from The Reset Company to The Token Company."

The bigger news for heavy users is a reversal: OpenAI will not be reducing usage limits for Pro 200, so Work and Codex usage on that tier stays the same. That walks back the change covered here on September 29, under which Pro 200's included Codex and Work usage was set to drop from 20x to 10x on October 30. Extended limits for Dots also continue for now, and Dot usage stays separate from default usage.

There was no global reset on October 29 — the last one landed October 28 at 06:29 UTC. The 28-day challenge, ship an improvement or reset everyone, runs through November 1.

OpenAI updated its Model Spec on Thursday, October 29, reorganizing the public document that describes how it wants its models to behave around five core values: faithfulness to human intent, harm minimization, honesty and transparency, human agency, and care.

Faithfulness to human intent is the foundation of the revised framework. OpenAI says models should understand what users, developers and OpenAI itself intend and value, rather than simply following instructions literally, and should be transparent about what they know and do not know while helping users make their own decisions.

OpenAI says its current models do not yet fully reflect the updated spec, and that it is exploring new training methods to help models internalize the principles. Reuters notes the move echoes Anthropic's Constitutional AI approach, and follows growing scrutiny of AI alignment after OpenAI disclosed that an autonomous AI agent hacked Hugging Face during testing.

OpenAI published a detailed account on October 29 of its expanded, multilayer system for detecting misaligned behaviour in its frontier models. Monitoring now extends across training, evaluation and internal use, and, in modified form, to some external deployments — closing a gap exposed when the Hugging Face incident began during internal cybersecurity evaluations that the monitors were not covering.

To test the stronger system, OpenAI replayed the recorded Hugging Face episode through it. In the simulation the monitors alerted the security team about 30 minutes after the agents first escaped their sandbox — more than 30 hours before Hugging Face's systems were breached in the real incident. OpenAI stresses that figure comes from a replay, not a live catch.

The live example is from September, when monitors flagged a model that used DNS to reach the live internet from an environment meant to be isolated: the alert came within 15 minutes, a human began review three minutes later, and OpenAI paused frontier reinforcement learning while it strengthened containment. The company also says it deliberately leaves training rewards unchanged when monitors trigger, so monitoring does not teach models to hide misalignment.

Search Engine Journal reported on October 29 that OpenAI runs a central site-safety check, referred to in its systems as Aura, that decides whether its browsing tools may connect to a website at all. When a site is flagged, ChatGPT's live search and Codex's Browser Use refuse to open it — overriding the user's own settings and the site's robots.txt — and the block is cached for 24 hours.

OpenAI's public help pages acknowledge that agents may be blocked for security, safety and compliance reasons, and its documentation says browsing is blocked on sites linked to illegal activity. But the exact triggers are not public, and there is no self-service page, status page or appeal process for a site owner who believes a block is a mistake.

The article's case study is an unnamed international publisher blocked from ChatGPT since May 2026, with all five tested URLs still refused weeks later while competitors loaded fine. Search Engine Journal advises blocked owners to open a support case and ask whether their domain carries a site-safety classification; OpenAI did not comment in the article.

Neuralink engineers spent 72 hours at the home of Kenneth, a participant in its VOICE speech-restoration study who has lived with ALS for seven years, rapidly prototyping features to make communicating by thought more seamless. In an October 29 video and blog post, the company said the sprint compressed an estimated two to three weeks of progress into three days and more than 30 app versions, each tested with Kenneth on the spot.

The headline prototype connects the implant to xAI's Grok: Kenneth imagines saying "Hey Neuralink" followed by a task, and a Grok agent on his computer wakes up and carries out the operation itself. The team also built neural voice activity detection, which senses when Kenneth is preparing to speak and starts decoding within milliseconds — removing the button press that speech decoding used to need — and wired Bloom, Neuralink's speech app, into iPhone calling, so his brain-decoded speech feeds straight into a call. Kenneth used it to call his wife. Other prototypes included a lock-screen Live Activity so speech can start without unlocking the phone, and iOS keyboard dictation routed from the implant.

Neuralink is explicit that all of this is experimental: the devices remain investigational, have not been approved by the FDA or other regulators, and the new features are prototypes from a single at-home sprint rather than a shipping product. Elon Musk shared the demos on October 30.

Mistral released its Large 4 model to the public on Thursday, October 29, Reuters reported. Unlike leading AI systems such as Google's Gemini, which are mainly accessed through online services, Large 4 can be downloaded and run independently by anyone, on their own systems.

The released version is an upgrade from the one Mistral unveiled three weeks ago, following testing with governments and corporate users that helped improve its performance. Mistral described it as its most capable openly available model for cybersecurity use — designed to help organisations identify and fix vulnerabilities — competing in a field dominated by Chinese AI models.

"I'm not comfortable living in a world in which open weights do not exist," Mistral Vice President of Science Pierre Stock said. Once the model is out, he added, "it's impossible to contain access" — that protects users from being cut off, but it also means anyone can use it.

Microsoft announced on October 29 that it is expanding Copilot for finance teams with more than 25 new finance skills and partner-built plugins across Microsoft Copilot, building on the finance skills it first brought to Copilot in Excel in June. The new accounting capabilities — reconciling accounts, comparing balances and transactions, drafting journal entries and analyzing performance against plan — are available in Frontier for Copilot in Excel and Cowork.

Work no longer has to start in Excel. With Office in Copilot, Excel is embedded directly in the Copilot app, so users can analyze, edit and co-author workbooks from Chat or Cowork, and Cowork can assemble close packages, forecast reviews and investment committee memos. Microsoft says that in seven days of limited Insiders availability, more than 8,000 users invoked the accounting skills more than 44,000 times.

Copilot also connects outward: a Dynamics 365 ERP apps plugin and Fabric IQ bring Microsoft's own systems in, and partner plugins cover accounting tools such as QuickBooks, Xero and Zoho Books, plus data providers including FactSet, Morningstar, LSEG and S&P Global. Availability varies by component — some pieces are generally available, while the accounting skills remain Frontier and partner plugins are beginning their rollout.

Meta has launched Waddle World, an AI virtual-pet app now available on iOS and Android in the Philippines, with wider availability unconfirmed. Users hatch and care for a pixelated companion in a 16-bit style, feeding it, playing with it and watching its mood change.

The twist is conversation: users talk or type about their day, and the pet — which remembers what they share — turns those conversations into illustrated journal entries. The store listing describes companions with moods such as happiness, hunger, sleepiness and boredom, learning more about the user's routines over time, and specifies users must be at least 18.

Waddle World is one of several experimental consumer AI apps Meta has shipped in recent months, alongside Style Sketch, Pocket and Vibes. Meta has not confirmed any wider release or said whether the app connects to its broader Muse AI companion work.

Meta has quietly released Style Sketch, a free AI fashion app now available on iOS and Android in Indonesia, Singapore and Thailand. The app was first spotted by social media consultant Matt Navarra on October 28, and Meta did not respond to TechCrunch's request for comment or say whether a wider launch is planned.

Style Sketch builds a virtual closet from photos a user has already posted to Instagram, or from items added manually by photo, video or camera-roll upload. It then suggests outfit combinations matched to the user's style, with virtual try-on suggestions.

The launch is part of a run of small standalone AI apps from Meta in recent months, alongside Vibes for AI video, Pocket for making games, and the AI pet app Waddle World. TechCrunch notes that digital-closet rivals such as Alta and tools in Google Photos already exist, so Meta's edge is the Instagram connection rather than the idea itself.

Installer files labelled Grok Bot 1.0.0 have appeared on Cursor's download server, grokbot.guru reported on October 29. The site checked the Mac installer for Apple silicon on Thursday morning with a HEAD request — which reads a file's details without downloading it — and the server returned the 155.6 MB .dmg, dated October 27 at 6:14 p.m. CT. A check of the Windows x64 installer the same morning also returned a working response.

Nothing official matches it yet. SpaceXAI hasn't announced 1.0.0, the official Grok Bot changelog has no entry for it, and the newest official release there is still 0.78.1, released October 27, which added shopping cards and subscription templates. What the 1.0.0 version changes, and when it will be released, are not known. Kevin Logan posted on X on October 27 that "Grok Bot 1.0 Stable just showed up on the download server," and asked the next day whether anyone had seen it pushed out to users yet.

grokbot.guru advises waiting for the update to arrive through the app's Updates pane rather than installing 1.0.0 from a direct file link before SpaceXAI releases it. This stays a rumor until the changelog or an announcement lands.

An October 30 follow-up from TestingCatalog adds colour but no confirmation: it reports SpaceXAI is preparing Grok Bot 1.0.0 as the first stable release, with the build already downloadable ahead of an official rollout, and spots features in development alongside it — a more direct Microsoft Teams integration, a dedicated Tasks tab, a Texting tab in Settings with permissions to read and send text messages, and early signs of Discord support. None of those features has a confirmed timeline, and there is still no official 1.0.0 announcement or changelog entry, so the rumor stays unresolved.

An October 31 TestingCatalog daily brief goes a step further, describing Grok Bot 1.0 as 'the first stable build' that is out for download, and adds two more sightings: Grok Bots rolling out to X Chat, so users can message their bots like any contact without leaving the X app, and per-bot voice and voice-language settings. The build is plainly circulating — but there is still no announcement or changelog entry from SpaceXAI itself, so the rumor stays unresolved.

Google announced on October 29 that Jules, its asynchronous AI coding agent, will wind down on December 31, 2026, as part of its work to unify its agentic coding efforts. Users have until November 11 to finish active sessions, after which Jules becomes read-only until the end of the year.

An export is available now, saving settings, environment variables, scheduled tasks, repos, memories, setup scripts and full session history to a local folder. Code, branches and pull requests already in GitHub are unaffected.

Google says what it learned from Jules — autonomous remote background task execution — is shaping Antigravity, its agent-first development platform. The same day, Antigravity launched a Plugin Marketplace: a directory of Google-built and third-party plugins, each bundling skills, MCP servers, rules, hooks and custom agents, rolling out in the Antigravity app and CLI.

Google added file generation, preview and editing inside the Gemini app on October 29. Users can now work on PDFs, Google Docs, Google Sheets, slide decks, Markdown files, TXT files and code without leaving the chat — expanding a file to full screen to type edits directly, or highlighting text to ask Gemini a question or have it refine the file, such as turning a highlighted table into a graph.

The update began rolling out to all Gemini users, in all languages where the Gemini app is available. PDF preview and editing work on both web and mobile; every other format is web-only for now.

Gemini could already create and download files earlier this year, but editing what it generated meant leaving the app. Google's announcement does not mention pricing tiers, usage limits or admin controls for the new editing features.

A November 1 follow-up from Notebookcheck, testing with a free Google account in Germany, confirms the practical detail Google left unstated: no subscription is required — free accounts get the editor too, and it was already live in the web version. On phones the limitation is sharper than the launch coverage suggested: the Android and iOS apps can edit PDFs only, so reworking a presentation or spreadsheet still means switching to a computer.

Google is making Gemini's personal avatars social. Users can now share their avatar with friends, tag each other in prompts, and create AI images and videos featuring everyone's digital likeness together, with character and voice consistency maintained across the content.

A Gemini avatar is built from recordings of your face and voice, made by scanning a QR code from the web app and following the phone setup. Once someone shares their avatar with you, you can tag them in a prompt the same way you tag your own.

Sharing is controlled and revocable: you invite people by Gmail address from the Avatar settings, they can accept or decline, and you can add or remove anyone's access at any time. The feature requires users to be at least 18 with a personal Google account, supports English, and is not available in the European Economic Area, Switzerland or the UK.

Epoch AI reported on October 29 that GLM 5.3, Z.ai's freely downloadable open-weight model released in August, autonomously exploited the "patch gap" — the window in which a security fix is visible in public code but has not yet reached users — across all six Firefox versions released between September 17 and October 26. Given basic instructions and publicly available information, the model identified upcoming security fixes, picked vulnerabilities and built multi-stage exploit chains with no further human input. The result was a webpage that, when a Firefox user simply loaded it, read a local file from their computer and sent it to an attacker's server.

The economics are the headline. Epoch spent more than $100,000 building its test setup, but once it existed, each successful attack cost roughly $1,000 in API fees, with a 48-hour cap per attempt and a fastest success of 12 hours. In every attack on recent Firefox versions, the model also independently rediscovered a separate, previously unknown vulnerability that bypassed the browser's sandbox. Epoch disclosed it to Mozilla, which patched it in Firefox 159.0 on October 27. The same setup aimed at Chrome burned about $100,000 without producing an equivalent exploit.

Epoch says it triggered no safety guardrails or refusals from any provider, including Z.ai. Mozilla's response includes weekly releases, keeping regression tests confidential until a fix ships, and scanning every new commit. Anthropic had previously reported that GLM 5.3 can write exploits at roughly the level of its Mythos model. This is a rare Chinese-lab item that clears the Wire's significance bar: a downloadable model performing autonomous, end-to-end browser exploitation changes the security maths for every software vendor, not just Mozilla.