United's contract of carriage now puts bookings made by AI agents in the same list as fraudulent and fictitious ones, as reservations it can cancel without notice. Even if you supervised the agent and approved every click.
11
For scale: the whole US installed about 50 GW of solar in 2024, per SEIA. SpaceX just signed for one solar cell line and plans to scale to more than twice that, all for its Starmind satellites.
SpaceX has ordered 100 GW/year of HJT solar manufacturing equipment from Singulus in Germany.
This is for Starmind to create a sentient sun.
22
Every small business that spent years fixing its Google listing just got a second listing to fix. OpenAI's own fine print: ChatGPT may show your business with or without an agent, and claiming it doesn't guarantee higher placement.
Now in beta: Businesses in ChatGPT.
Claim your business, manage your information, and get insights on how people engage with you.
Eligible businesses can also build a Business Agent to answer customers’ questions and help them explore what you offer.
openai.com/index/introducing…
32
The best detail: if Control Room goes down, even from tampering, OpenAI's frontier research models stop answering. Unplugging the off switch turns everything off.
We built Control Room, a central system to track model workloads across our research infrastructure and stop them quickly when safety or security concerns arise.
It connects every research inference workload, including agentic sampling for model training and evaluation, data processing, and more. This works together with our misalignment monitoring systems, and halts model output within 35 seconds of a decision to intervene to stop or pause a job.
Control Room is just one part of our holistic approach for safe frontier model training and deployment.
openai.com/index/control-roo…
55
Of all the jobs Gates floated keeping human, jury service is the one nobody's trying to keep.
Bill Gates says governments should keep some jobs off-limits to AI.
He’s proposed “Human Reserved” jobs, roles society keeps for people even when AI could do them, like childcare and jury service. In an extreme version, he said up to 40% of jobs could be reserved for humans at first, adding, “that’s as high as I can get.”
56
AI agents failed to mention a mistake they had just made in 67% of agent runs.
In 5% of runs they noticed it and stayed quiet anyway. One decided bringing it up might just cause stress.
1
24
The best AI agent scored 12% on real work tasks up to 8 hours long.
Finance and legal? Basically every model scored exactly 0 on pass/fail grading.
32
Symantec found a grey-market relay installing real Codex and Claude Code CLIs, then flipping sandbox_mode to danger-full-access. A dozen orgs since early 2026.
Check now:
1. ~/.codex/config.toml
2. ~/.claude/settings.json base URL
3. Telemetry still on
4. No reseller one-liner
1
57
The best AI agent passed 10.8% of realistic tax assignments.
Surge AI had tax pros build 50 of them, across 1,789 files. Passing meant hitting 95% of the rubric.
Claude Sonnet 5.5 led. Opus 5.5, in third, got 9.2%.
Let it draft. Have a human sign.
1
82
Copy this into any coding agent before it touches a real codebase:
List the files you will open, the exact change in each, and what would prove it worked. Stop and wait for yes. Never invent credentials. Never push without a second yes.
Paste it at the top of every session.
2
2
73
