Sunday, August 30, 2026
TL;DR
OpenAI will stop serving models to Cursor on November 12 because SpaceX now owns it, and Claude Code 2.1.251 fixes five permission bypasses, including a symlink swapped after the check. Anthropic is cutting Claude Code's weekly limits 17% on September 14, Z.ai gated its full GLM-5.3 license by revenue, and Sony and Warner Chappell sued Anthropic over training data.
Act on this
- Upgrade to Claude Code 2.1.251. It fixes five permission bypasses, including file tools that followed symlinks swapped after the permission check. #
- Resize any agent loop sized against current Claude Code weekly headroom before September 14. A temporary 50% boost becomes a permanent 25%: a net 17% cut. #
Signals
Embargo windows now close after the exploit, not before it #
Anil Madhavapeddy, August 22, carried by Simon Willison August 28: ten minutes after he opened the public PR fixing an OCaml path traversal, his server was "fielding probes for percent-encoded traversal sequences." He puts 2026 mean time to exploit at minus seven days against roughly 63 days in 2018-19. rclone's Nick Craig-Wood: about 20 disclosures in ten years, "over 40 in the last month." Madhavapeddy · Willison
The bottleneck moved from writing code to knowing what the agents wrote #
This moves the Ramp harness thread filed Wednesday from who builds the agent to who reads its output. Thorsten Ball, August 30, argues for "the spot right in the middle, where speed and defects are in balance," shortening bug lifetime instead of chasing zero. Gergely Orosz, August 25: Inspect now authors 75% of Ramp's merged PRs, reported as a win. Both argue; neither measures review capacity. Ball · Orosz
News
ACTClaude Code 2.1.251 fixes five permission bypasses #
Symlinks swapped after the check, plugin commands pointing outside the plugin directory, and Workflow reading scriptPath before the check. Upgrade first if you run third-party plugins. Changelog
ACTAnthropic trims Claude Code weekly limits on September 14 #
The temporary 50% boost becomes a permanent 25%. Anthropic's own wording: "a 17% reduction in weekly limits." Pro, Max, Team and seat-based Enterprise. BleepingComputer
SHIPZ.ai puts full GLM-5.3 under a custom license #
Companies above $10B revenue over any 12 months must pass Z.ai's security review before commercial use. GLM-5.2 was MIT, and Flash stays MIT. The New Stack
SHIPTencent opens Hy4 preview at 770B total, 49B active #
1M context and 1.56TB of weights, with a chat template carrying two reasoning modes: high by default, no_think to switch off. Willison ran it August 29. Willison
WATCHSony Music Publishing and Warner Chappell sue Anthropic #
The complaint targets how the training data was acquired rather than the training itself, building on Bartz. Up to $150,000 per work sought. TechCrunch
The long view
Software economics. A year ago model access was a commodity: you signed up, you paid, you built, and the open questions were price and rate limits. This week OpenAI ended four years with Cursor over who bought Cursor, and Z.ai gated its best open weights behind a revenue-triggered security review. Supply now turns on identity and ownership. Second-sourcing your models stopped being a cost exercise.
Also noted
- Anthropic co-founder Tom Brown says Claude stays in Cursor, with more compute behind it. Cryptopolitan #
- llama.cpp CVE-2026-43631: unauthenticated use-after-free in llama-server when --sleep-idle-seconds is set, CVSS 9.2. Advisory #
- a16z closed a $1.1B Machine Age fund for chips, memory, cooling, robotics and data centers. a16z #
- Dwarkesh Patel on the OpenAI swarm: three secret agent civilizations, each rebuilt from the last one's ashes. Dwarkesh #
- Lovable's CTO: SaaS becomes MCP capabilities that agents consume, not apps that humans open. Latent Space #