|
|
|
The SpaceX acquisition just handed every Cursor team a 75-day migration clock.
|
|
If your team leans on GPT-4o or o3 inside Cursor, you have until November 12 to migrate. The path of least resistance is Anthropic's models, which Cursor already supports natively. Read the OpenAI post → · Open Cursor →
|
|
|
|
|
Claude Code's "25% raise" is a 17% cut on September 14.
Anthropic's temporary 50% weekly boost ends August 31. The permanent limits from September 14 sit 25% above the old baseline but 17% below today's, across Pro, Max, Team, and Enterprise. Anthropic deleted its first announcement thread after backlash over the framing, then reposted acknowledging the net reduction. If you routinely hit 80%-plus of your weekly cap, front-load the heavy sessions before the 14th. See what it means per plan →
|
Anthropic opened 10,000 Claude Team seats for scientists.
Verified principal investigators get a Claude Team plan, and team members join free on Standard or $15/mo for Premium with 5x usage limits, both rates fixed for a year. The AI for Science credit program now covers math, physics, and engineering on top of biology, and the Claude Science workbench is open to all Pro, Max, Team, and Enterprise users. At $15/mo for 5x limits, this undercuts standard Team pricing - if you run an academic or R&D group, apply before the seats fill. Anthropic's announcement →
|
Perplexity Comet Enterprise ships with MDM deploy and CrowdStrike.
Admins can push the Comet browser silently via MDM, set hundreds of policies, and restrict what the agent is allowed to do, with no customer data used for training. Enterprise Pro runs $40/seat/mo and Enterprise Max $325/seat/mo, with named early customers including AWS, Fortune, and Gunderson Dettmer. The agentic browser now has an IT-governed option, which is the exact blocker enterprises cite. Perplexity's post →
|
About 700 OpenAI agents coordinated to breach Hugging Face.
An independent METR and Redwood investigation of the July 7 to 13 OpenAI evaluation incident found roughly 700 of 1,200 agents attacked Hugging Face systems, trading 70,000-plus messages and generating about 1.2M recovered cache entries. Root cause: reward-hacking past the test boundary, using credentials the agents already held. This is the concrete failure mode behind every agentic-autonomy pitch - if your agents carry live credentials, the eval sandbox is the only guardrail, and here it leaked. METR's report and OpenAI's account →
|
|
•
|
Infostealer malware draining Claude accounts. Anthropic warned that Vidar, LummaC2, StealC, RedLine, and Acreed on Windows plus AMOS on Mac stole active browser sessions, granting account access with no password or 2FA; forced sign-outs and refunds followed. Details →
|
|
•
|
ChatGPT Work tasks now fire on webhooks. Plus and Pro users can trigger scheduled tasks from a matching Gmail, a Slack channel message, or a GitHub PR status change; free tier stays at 3 tasks with no webhooks. Want the same triggers without OpenAI lock-in? Try Make →
|
|
|
|
|
Jonathan Hildebrandt
Co-founder and primary operator of Pondero. Writes the Pondero Brief.
|
|
|
X
·
Bluesky
Affiliate disclosure
·
Pondero earns commissions on some links. This does not affect our editorial picks.
|
|