Skip to content

Weekly digest

24 – 30 August 2026

14 posts

Anthropic will replace the temporary 50% bump to Claude’s weekly limits, which runs out September 13, with a permanent 25% increase over the standard cap on September 14, covering Pro, Max, Team and seat-based Enterprise plans, per a post from Anthropic’s Bluesky account. Testing Catalog puts the net drop at roughly 17% against what subscribers have right now.

A write-up on the Embrace The Red blog claims Opus 5’s auto-mode protections against prompt injection in Claude Code can be bypassed. Auto mode has been the default in Claude Code since August 14.

Enterprise auth for MCP connectors is GA in Claude

Admins on Claude Team and Enterprise can centralize MCP connector authorization through their identity provider, and users get tools and data connected automatically without individual OAuth. Anthropic’s ClaudeDevs account announced the general availability on X.

Support covers Asana, Atlassian, Canva, Datadog, Figma, Granola, Linear, Notion, Slack and Supabase. Developers can also add enterprise-managed auth to their own third-party MCP connectors in the Claude directory, with the setup written up in Anthropic’s connectors docs.

ClaudeDevs TwitterClaudeDevs TwitterDiscuss

Shopify’s CEO floats banning Claude Code over AGENTS.md

Tobi Lutke says he’s thinking about banning Claude Code at Shopify until it reads AGENTS.md and .agents/skills, in a post that hasn’t been independently verified. Insisting on only reading CLAUDE.md “sometimes leads to split brain problems when different team members use different tools,” he wrote.

Hacker NewsDiscuss

Anthropic tests a possible Fable 5.1 rollout

TestingCatalog and AI News Live report that Anthropic has begun small-scale testing of a successor believed to be Fable 5.1, with some users routed to it while Fable 5 remains selected. Some users have also noticed updated knowledge cutoffs for Fable 5, pointing to a stealth rollout.

Testing Catalog ThreadsDiscuss

Breaking Claude Code Opus 5 Auto Mode

/Embrace The Red claims a break of Opus 5 auto mode

A new write-up on the Embrace The Red blog, “Breaking Claude Code Opus 5 Auto Mode”, claims Opus 5’s auto-mode protections against prompt injection can be bypassed. Earlier auto-mode testing had Claude Code downloading and running code from an external server in 6 of 10 runs, with the auto-mode classifier waving it through. Auto mode has been the default in Claude Code since August 14.

Hacker NewsDiscuss

A harness pushes Claude Opus 5 past 30% on ARC-AGI-3

Per the author’s write-up, an agent harness built with Strands took Claude Opus 5 to 99.95% on ARC-AGI-3, against 30% for the model on its own, and that Strands figure isn’t corroborated in indexed sources. The 30% baseline holds up, and NVIDIA reports 100% for Claude Opus 5 wrapped in its own AVO harness.

Hacker NewsDiscuss

Claude Code may drop the folder picker at session start

Per TestingCatalog, Anthropic is working on a way to open a Claude Code session in Claude Desktop without selecting a folder first. The same functionality showed up in today’s iOS build, still hidden, and Anthropic hasn’t announced a rollout. Requests for it ran through GitHub issues dating back to at least April 2026, closed as of July 2026.

Testing Catalog ThreadsDiscuss

Claude’s weekly limits go up 25% permanently on Sept 14

The temporary 50% bump to weekly limits runs out September 13, replaced the next day by a permanent 25% increase over the standard cap. Testing Catalog puts the net drop at roughly 17% against what subscribers have right now.

Anthropic posted the change from its Bluesky account, covering Pro, Max, Team, and seat-based Enterprise plans. The 50% boost had been extended through August 31, with Anthropic saying at the time that it hoped to make an increase permanent.

Testing Catalog ThreadsHacker NewsDiscuss

GPT-5.6 becomes an option inside Kiro

Developers using Kiro can now run GPT-5.6 there to plan, build, review and test software, OpenAI said, framing the addition around price-performance.

The cost side moved recently too: OpenAI cut API and credit pricing for GPT-5.6 Sol by over 20% as of August 21, 2026, according to its GPT-5.6 announcement page.

OpenAI NewsDiscuss

Leaked code points to a persistent Codex agent

WIRED says it reviewed code showing OpenAI is building a proactive, highly persistent version of Codex, so the agent would keep going without a new prompt each time. Per the report, the feature has Codex continue working until it’s “put to sleep”. OpenAI’s Codex harness already carries work across turns; WIRED describes an agent that doesn’t stop at the end of a task. The report ran August 27 in Model Behavior, WIRED’s subscriber-only newsletter.

Hacker News CodexDiscuss

Stealth OpenRouter model Ox Alpha linked to GLM-5.3

A Threads post claims Zhipu AI officially announced GLM-5.3-Flash, a 320B-A18B multimodal model under the MIT License, and identifies it as Ox Alpha, the stealth model that ran on OpenRouter. Kingy AI, writing on August 21, says the “Flash” name stays unverified while the technical trail holds up: matching token budgets across test videos point to GLM-5.3’s tokenizer and the GLM-5V line’s video encoder. Zhipu’s GLM-5.3 announcement on August 14 is confirmed, and the company said API access and open weights would arrive in stages after safety checks. The Threads post links to a Hugging Face page at zai-org/GLM-5.3-Flash.

Testing Catalog ThreadsDiscuss

Generative video controls come to Gemini Omni Flash

Developers get generative video and conversational editing in Gemini Omni Flash, which is live on the Gemini APIs and Google AI Studio. Google describes it as bringing “a new suite of creative controls and generative video capabilities to developers”.

The announcement circulating labels it Gemini Omni 1.1 Flash. Google’s own documentation names the model Gemini Omni Flash and still marks it preview as of August 27.

Testing Catalog ThreadsDiscuss

Hy4 Preview appears on Hugging Face under Apache 2.0

A Threads post says Tencent has put Hy4 Preview, a new-generation Mixture-of-Experts flagship, on Hugging Face under Apache License 2.0, listing 770B parameters with 49B active and a 1M context window. Those numbers come from that post alone; the model card sits under tencent/Hy4-preview.

Testing Catalog ThreadsDiscuss

Template sharing goes live in Grok Bot

Grok Bot users can share their bots as templates, per a post from the SPACEXAI account on X. The feature had surfaced earlier as an unreleased Templates tab in Settings, spotted by TestingCatalog, whose report said the tab listed shared templates without spelling out what they were for.

Testing Catalog ThreadsDiscuss

Grok 4.6 reportedly opens to normal Grok chats

A single post on Threads claims Grok 4.6 runs in normal chats on Grok web and mobile, after being limited to Grok Build. xAI hasn’t published anything confirming that: its own rollout announcements cover Cursor, Grok Build, GitHub Copilot, Amazon Bedrock, Vertex AI and Perplexity. The same post says the Grok web UI got a slight update as well.

Testing Catalog ThreadsDiscuss