On mvanemmerik.com, Nov 16: NFL Week 10 chalk talk: Detroit dazzles, Cincinnati takes the North
vanemmerik.aiClaude, ChatGPT and Grok news, twice a day

Monday, November 9, 2026

Anthropic releases Claude Fable 5.2 and Mythos 5.2, faster and cheaper than 5.1

Fable 5.2 reaches all API platforms at $8/$40 per million tokens, while Mythos 5.2 stays with vetted researchers. Elsewhere at Anthropic, Claude Duet gained a Sessions view for Claude Code, researchers found secrets leaking to GitHub through Claude Code's saved approvals, and South Korea's privacy regulator is reviewing an exposure of about 100 business accounts. xAI began watermarking Grok API text by default, and Musk promised overnight fixes after Grok Bot subscribers said allowances drained in hours. OpenAI was quiet.

Monday, November 9, 2026

ClaudeModel

Anthropic releases Claude Fable 5.2 and Mythos 5.2, up to 30% faster than 5.1 and priced at $8/$40 per million tokens

Anthropic released Claude Fable 5.2 on Nov. 9 across the Claude Platform API, AWS, Google Cloud and Microsoft Azure, alongside Claude Mythos 5.2, the same underlying model with lighter cyber and biology safeguards that stays limited to approved security and life sciences researchers and to Claude Security for Enterprise customers. Both cost $8 per million input tokens and $40 per million output, down from $10 and $50 for Fable 5.1, with an optional fast mode at $16 and $80 that Anthropic says generates tokens up to 2.5 times faster. Anthropic says outputs are up to 30% faster than 5.1, Fable 5.2 trips about 55% fewer safeguard blocks, and Pro plans and standard Team seats can spend up to half their weekly usage on Fable.

Why it matters: It is Anthropic's new top model for general use and cuts its flagship API price by a fifth, slightly below OpenAI's GPT-6 Astra list price.

Try it: In the API use the model ID claude-fable-5-2; in Claude apps and Claude Code pick Fable 5.2 from the model menu on an eligible paid plan.

ClaudeClaude APIClaude CodeHappened Nov 9, 2026 · Sources: Anthropic: Introducing Claude Fable 5.2 and Claude Mythos 5.2 · Claude Platform Docs: Pricing · Fortune

ClaudeSafety

Researchers find 2,539 secrets on GitHub saved by Claude Code's "don't ask again" command approvalsReported

Security firm Apiiro says that when a Claude Code user approves a command with "Yes, and don't ask again", the full command, including any password or token typed into it, is saved in the project's .claude/settings.local.json, which often ends up committed to public repos. Searching GitHub, it counted 2,539 secrets across 1,994 repositories, including database passwords, SSH logins and AWS keys, The Hacker News reports; Anthropic has not published a fix or statement, and an earlier bug report about the behavior was closed by a bot.

Why it matters: Anyone who has committed that file may have published live credentials that need rotating.

Try it: Check your repos for .claude/settings.local.json, rotate any secret it ever held, untrack it with git rm --cached and add it to .gitignore.

Claude CodeHappened Nov 7, 2026 · Sources: The Hacker News

GrokPolicy

xAI starts watermarking text from Grok 4.7 and six other API models by default, citing the EU AI Act

Since Nov. 7, text from grok-4.7, grok-4.6, grok-4.5, grok-4.3 and three grok-4.20 variants on every xAI API endpoint carries an invisible statistical watermark, including structured outputs and tool-call arguments, according to xAI's developer docs. Responses that carry it return an x-watermark: true header, detection is available only to approved applicants, and team admins can opt out per model only after confirming a legal exception applies; xAI says no code changes are needed.

Why it matters: Developers building on Grok now ship watermarked text by default, ahead of EU AI Act marking rules that take effect for older systems in December.

xAI APIGrokHappened Nov 7, 2026 · Sources: xAI Docs: Output Watermarking · Emergent

ClaudeSafety

Anthropic reported to South Korea's privacy regulator that contact and plan data for about 100 Korean business accounts was exposedReported

South Korea's Personal Information Protection Commission said on Nov. 9 that it is doing a preliminary review, short of a formal investigation, after Anthropic reported on Oct. 31 that account IDs, names, email addresses and subscription plans tied to roughly 100 Korean enterprise Claude accounts were exposed, Seoul Economic Daily reports. The reports link the exposure to security startup Cathedral's late-October access to Anthropic's business systems through a flaw in third-party software; an Anthropic staff member said on X that prompts, outputs and payment data were not accessed. Anthropic hasn't published its own statement.

Why it matters: It is the first regulator review tied to the Cathedral access, weeks before Anthropic's planned IPO.

AnthropicClaudeHappened Nov 9, 2026 · Sources: Seoul Economic Daily · Phemex News

ClaudeFeature

Claude Duet adds Claude Code sessions, so you can follow, message and answer them from your phone

This weekend's update to the Claude Duet iOS app (version 1.261108.0) adds a Sessions view where users can follow their Claude Code sessions, send them messages and answer their questions, according to Anthropic's App Store release notes. Duet remains a US research preview for Pro, Max and Ultra subscribers who have been let in from the waitlist; Anthropic hasn't published a separate announcement or said whether the Android app has the feature yet.

Why it matters: It links Anthropic's always-on agent app to its coding agent, so developers can check on long-running coding work from the same app that handles their other tasks.

Try it: Update Claude Duet on iPhone and open Sessions; Claude Code sessions tied to the same Claude account should appear there.

Claude DuetClaude CodeHappened Nov 8, 2026 · Sources: Claude Duet on the App Store (release notes) · Claude Help Center: Get started with Claude Duet

GrokFeature

Update: Musk says Grok Bot fixes are going in overnight after subscribers say paid allowances run out within hoursReported

After a subscriber posted on Nov. 7 that a $200 Grok Bot plan that once lasted a week of heavy use ran out in 14 hours of moderate work, Elon Musk said on X on Nov. 8 that token-efficiency fixes were underway and, that evening, that users should see a dramatic improvement by Monday. He didn't say what the fixes change, and xAI hasn't announced higher weekly limits, a usage reset or price changes, Tesla North reports.

Why it matters: Usage allowances decide how much work an always-on agent can do for its monthly price, and subscribers say theirs shrank sharply this month.

Grok BotHappened Nov 8, 2026 · Sources: Tesla North (Nov. 8, evening) · Tesla North (Nov. 8)