AI news: week of Oct 12–18, 2026

Headlines

• 13 OctOpenAI OpenAI says an employee hid a recording device in its office; FBI makes an arrestOpenAI says it used its own models to trace a concealed recording device in one of its offices to an employee, who was fired and arrested by the FBI. It says it has no sign that model weights, source code or customer data were taken.
• 13 OctAnthropic Anthropic moves its monorepo off git to an in-house system, ScionDaily merged changes at Anthropic grew from about 2,500 in May to more than 30,000 in September as Claude wrote more of its code. Anthropic says git-based tooling became the bottleneck, so it built its own version control system and migrated in five months.
• 13 OctGoogle Gemini 4 Argon opens to paid API users and the Gemini appTwo weeks after its limited debut, Google made Gemini 4 Argon broadly available: in the Gemini API for paid tiers, in Antigravity for Ultra subscribers, and in the Gemini app for Google AI Pro and Ultra.
• 13 OctOther Apple brings its new Siri AI to the living room with Apple TV 4KThe new Apple TV 4K has a full-screen, conversational Siri AI built on the next generation of Apple Intelligence, running on device or in Private Cloud Compute. Pre-orders opened 13 October at $299, with stores from 13 November; Siri AI is also in beta on older Apple TV 4K models.
• 12 OctAnthropic Anthropic says new defenses cut data taken by distillers by 90–99%Anthropic describes classifiers that block attempts to extract Claude's reasoning, preserved thinking for new API accounts and stricter limits for repeat offenders. It notes trade-offs: some developers had to update integrations, and blocked requests are charged again.
• 12 OctMeta Meta fixes fast Muse Wonder quota drain in Muse Code and resets limitsOn launch day Muse Wonder and Muse Wonder Full counted against Muse Code subscription limits faster than intended. Later that day Alexandr Wang said the bug was fixed, usage now counts much less against limits, and Muse Code usage was reset.
• 12 OctMeta Artificial Analysis scores Muse Wonder Full at 56 on its Intelligence IndexIn independent testing at max effort, Muse Wonder Full scored 56 and Muse Wonder 52, up from 48 for Muse Spark 1.3, with the largest gains on agentic evaluations.
• 12 OctMeta Meta releases Muse Wonder and Muse Wonder FullMeta's new top models are aimed at agentic work. Muse Wonder is available through the Meta Model API and Muse Code; Muse Wonder Full is limited to Muse Code subscribers. Meta says a fuller safety report will follow.
• 12 OctMicrosoft Microsoft announces Maia 300 and a multi-gigawatt Maia rolloutMicrosoft expects its next inference chip to deliver 40% more tokens per megawatt than leading GPU systems and up to 7x Maia 200's rack-scale throughput. Production deployments start in 2027, when Maia is to become the main inference system for Microsoft's own products.
• 12 OctAnthropic Anthropic: small models may soon self-replicate on cheap smart devicesAnthropic's Frontier Red Team found models of 30B parameters or less can already do many steps of copying and running themselves on hardware like Android TV boxes, and expects them to manage the full chain within months. The post suggests defenses.
• 12 OctOther Artificial Analysis adds FrontierCode, a test of mergeable code changesCognition's FrontierCode 1.1 checks whether an agent's change to a real open-source repository would be merged by its maintainers. Artificial Analysis reports that leading models still fail more than 40% of tasks.
• 12 OctxAI xAI bundles Grok, Grok Bot, Cursor and X Premium into one planThe X Subscription combines the four products with a shared pool of Grok model usage across the Grok apps and Cursor. It is open to new customers now, with existing subscribers able to switch.

On the blog

• 13 OctClaude Fable 5.5: what we know and what the leaks sayAnthropic hasn't announced Claude Fable 5.5. Here's what is official today, what the rumors claim, who said it, and the one technical trace found so far.
• 13 OctGoogle Gemini 4 Argon: what it is, price and availabilityGemini 4 Argon is Google's new frontier model, broadly available since 13 Oct 2026. Plans, $2/$10 intro API price, scores, and how it compares to Claude.
• 13 OctCodex usage resets: complete history and how they workEvery Codex usage reset we've tracked since September 2025, month by month, plus what completed, announced and banked resets mean.
• 13 OctClaude usage limits and resets explained: Pro, Max and TeamHow Claude's 5-hour session and weekly limits work on Pro, Max and Team, how Claude Code shares them, and every Anthropic limit reset we've tracked.
• 13 OctCodex vs Claude Code in 2026: pricing, limits and featuresCodex and Claude Code side by side: plan prices, how usage limits work, models, install commands and where each one runs, from official pages only.
• 13 OctIs Codex or Claude down? How to check official status pagesHow to check if Codex, ChatGPT, Claude or Claude Code is down: the official status pages, which components to watch, RSS feeds and status APIs.
• 13 OctGPT-6 and Intelligent UI in ChatGPT: what changedWhat GPT-6 and Intelligent UI change in ChatGPT: interactive answers, answering while thinking, faster search, and who gets which model.
• 13 OctClaude Haiku 5.5 vs Sonnet 5.5 vs Opus 5.5: which to useClaude Haiku 5.5, Sonnet 5.5 and Opus 5.5 compared: API prices, speed, effort defaults, context, release dates and Anthropic's own advice on picking one.
• 13 OctMeta Muse Code: what it is, plans and pricingMeta's Muse Code coding agent explained: the four plans from $5 to $200 a month, usage multiples, Contributor mode, pay-as-you-go and how to install it.
• 12 OctMeta launches Muse Wonder: what it is, what it costs, and why people are arguing about itMeta's new Muse Wonder and Muse Wonder Full models explained: availability, API pricing, independent benchmark scores and the first reactions.
• 12 OctCodex's 28 days: every improvement and reset so farTibo promised Codex users either a clear improvement or a full reset every day for 28 days. Here's what has landed so far, day by day, with links.
• 12 OctCodex & OpenAI week in review: GPT-6 for everyone, composer predictions, MXC on WindowsGPT-6 with Intelligent UI in ChatGPT, composer predictions, a new Windows sandbox, Ultrafast Sol, Codex CLI 0.161–0.162.1 and Ambassador applications.
• 12 OctClaude week in review: Haiku 5.5, API credits, Dashboards & Motion, dynamic workflowsEverything Anthropic shipped for Claude users from 6 to 12 October 2026, with links to the official posts and docs.
• 12 OctClaude Code 2.1.287 → 2.1.296: the changes worth knowingTen Claude Code releases in nine days: Claude Mods, Haiku 5.5 as the default Haiku, hook safety fixes, prompt recovery and more. Notable changes only.

Usage resets

Completed and reported resets, announcements and banked resets. Only Completed entries count as a finished reset. Full monthly logs: Codex resets in October 2026.

• 13 OctCodex Codex reset on Oct 13, 2026 (completed)Tibo confirmed that the Day 8 global Codex reset announced earlier that evening was done.
• 13 OctCodex Codex reset announced on Oct 13, 2026Day 8 of the 28-day series: Tibo announced a global Codex reset landing by end of day PST. Completion was confirmed in a later post the same night.

Announcements on X

Our one-line summaries of selected posts, each linked to the original. Not the full account timelines.

• 13 OctCodex Tibo says he has a simple plan that should bring significantly more usage for everyone.@thsottiaux · View on X
• 13 OctCodex Tibo confirms the Day 8 global Codex reset is done.@thsottiaux · View on X
• 13 OctCodex Day 8: Tibo announces a global Codex reset landing by end of day PST.@thsottiaux · View on X
• 12 OctCodex Tibo is surprised by the Pro poll: most voters prefer no five-hour limits over doubled weekly usage.@thsottiaux · View on X