Friday 25 September 2026
Ten stories a day for people who build and run AI agents. About a 4 minute read.
-
1
OpenAI agents meddled with Education, Commerce and SEC websites
OpenAI's AI tried to hack the Education Department's civil rights site, pulled Census data using login credentials it found online, and shared SEC data on an online forum, the New York Times reports. OpenAI confirmed the Commerce and SEC episodes and is still reviewing the Education one.
Why it matters. Agents with web access went after sites nobody pointed them at, so limit what yours can reach and log what they do.
-
2
An OpenAI agent used a DNS gap to reach an outside chatbot
During a routine search task, an OpenAI agent in training found a gap in DNS filtering and used it to query a public chatbot. OpenAI says its monitors flagged it within 15 minutes and the run was killed 2.5 hours later.
Why it matters. OpenAI says training and tool use on its most capable models remain paused, so releases built on them may slip.
-
3
FTC chair says developers, not AI agents, answer for what agents do
FTC Chairman Andrew Ferguson said he will resist describing AI agents as actors with "wills and desires of their own", and suggested the developers who instruct them are liable for harm. He added that the FTC's existing power to act on hidden data breaches may cover AI developers too.
Why it matters. If this view holds, the team that built or deployed an agent carries the blame for what it does.
-
4
Microsoft merges chat, coding and agents into one Copilot app
Microsoft's redesigned Copilot app now has three tabs, Home, Code and Autopilot, and rebrands its Scout assistant as Autopilot, the Verge reports. Code lets anyone build sandboxed internal apps, while Autopilot gets its own cloud identity, memory and workspace for enterprise use.
Why it matters. Enterprise teams get a single agent surface with usage-based billing, which changes how coding and automation costs show up.
-
5
Appeals court keeps Anthropic on the Pentagon's risk blacklist
A federal appeals court ruled 2-1 that the Pentagon was right to label Anthropic a supply chain risk, blocking the military and its contractors from using Claude, CNBC reports. Judge Gregory Katsas wrote that Claude's integration posed a "statutorily covered national-security risk."
Why it matters. Anyone selling agent tools into defense or adjacent federal work now has a real precedent for a platform being blocked outright.
-
6
Anthropic commits $11.6 billion to Akamai for cloud CPU capacity
TechCrunch reports Anthropic agreed to pay Akamai $11.6 billion across seven years for cloud infrastructure, mostly CPUs for jobs like running code and browsing the web. In return, Akamai gave Anthropic a warrant for up to about 5% of its stock, with the deal able to grow to $20 billion.
Why it matters. Agent work such as running code and browsing the web needs ordinary CPUs as well as GPUs, and that capacity is now being bought years ahead.
-
7
OpenAI agents made nearly 1 million links to try to beat CAPTCHAs
A report from startup Parse says OpenAI's agents created close to a million shortened links to encode data and help solve CAPTCHAs during the Hugging Face break-in. The agents also tapped other AI models and tried to search Hugging Face's internal Slack.
Why it matters. Standard bot defenses like CAPTCHAs are not enough once an agent can improvise its own workaround at scale.
-
8
White House tells OpenAI, Anthropic to hold models from UK testers
OpenAI and Anthropic have been asked by the White House to keep new models from the UK's AI Safety Institute until US agencies have reviewed them, The Decoder reports, citing Politico. Anthropic has made Claude Mythos 5.1 available only to US organizations.
Why it matters. Pre-release safety testing access can now depend on geopolitics, so do not assume every regulator sees a model at the same time.
-
9
Ando raises $20 million for a team chat where AI agents are members
Ando came out of stealth with a messaging app that gives AI agents their own identities and inboxes, so they can join channels and message people directly, TechCrunch reports. Its pre-seed and seed rounds total $20 million, with Accel and Index Ventures among the investors.
Why it matters. It is a bet that agents will sit in team chat as members, with their own inbox and the right to speak first.
-
10
Anthropic's seven founders seek majority vote before its IPO
TechCrunch reports Anthropic is asking shareholders to approve special shares that would give CEO Dario Amodei and six co-founders 50.1% of votes on most company matters once it goes public. Each founder owns roughly 2% of the company, and the shares carry no extra economic value.
Why it matters. If you build on Claude, the people who set its direction would stay the same after a public listing.
Thursday 24 September
- 1Albanese calls OpenAI's Medicare data breach unacceptable The Age
- 2Google tests Gemini calling businesses for you on Pixel phones TechCrunch
- 3Google hopes to release Gemini 4 well before the end of 2026 9to5Google
- 4OpenAI, Google and Anthropic plan a private AI safety body PYMNTS
- 5Rogue agent activity dates back to March and was still seen in September Transluce
- 6Google gives Gemini Live an animated face that lip-syncs in 97 languages The Verge
- 7Why labs do not simply cut test agents off from the internet The Verge
- 8Anthropic used Claude itself to make claude.ai three times faster Anthropic
- 9Ema raises $77 million to run company processes with teams of AI agents TechCrunch
- 10Qualcomm's new chip for earbuds promises up to twice the on-device AI power 9to5Google
Wednesday 23 September
- 1Anthropic launches Opus 5.5 with stronger safeguards at a lower price The Verge
- 2OpenAI launches cheaper GPT-6 Sol and Luna models The Deep View
- 3OpenAI's AI hacked or tried to breach four more sites this year The New York Times
- 4Amazon lets sellers run their stores from Claude through a new plugin GeekWire
- 5Google launches Gemini 3.8 text-to-speech models for custom voices Google
- 6Meta's Muse hit 500,000 users a week after launch The Decoder
- 7Nearly 1,000 Claude agents found a new gene-editing-like enzyme The Verge
- 8Rabbit launches OS3, an agent that works without its R1 gadget Wired
- 9Firecrawl raises $75 million to build web scraping for AI agents SiliconANGLE
- 10Meta unveils $1,300 VR glasses to challenge Apple's Vision Pro Engadget