Hello, dear TEA-mates! Here is what you need to know today.
1. ✂️ T-Mobile Snipped a Cable to Kick Out Chinese Hackers
New reporting from Bloomberg detailed how T-Mobile's cybersecurity team detected and expelled Chinese state-backed hackers from its network in 2024 during Salt Typhoon, a campaign that compromised hundreds of telecom and internet firms, including AT&T, Verizon, Viasat, Charter, and Windstream, to steal phone records on senior U.S. officials and presidential candidates. T-Mobile largely avoided a wide breach by catching the activity early. After months of fruitless searching, staff traced unusual traffic to a router belonging to another unnamed telecom. Cybersecurity chief Jeff Simon said he and three colleagues drove to a data center near T-Mobile's Bellevue, Washington headquarters, found the compromised box, and physically snipped the cable connecting it to the outside world with a pair of scissors. (Read More)
🫖 TEA For Thought: "Americans severely underestimate how much the CCP can intrude into infrastructure globally."
2. 🛡️ A Rulebook That Only Calls an Agent Successful If Every Step Was Allowed
A new arXiv paper, A Policy Algebra for Trust-Preserving Agentic AI Execution by Bhaskar Tripathi, Anurag Kumar, Ramendra Kumar, and Bhavesh Gadhe, argues that an AI agent completing a task is not enough if it got there through unauthorized data access, widened delegated authority, unapproved side effects, or unrecoverable budget spend. The authors define reliable capability as a path property: every action event must stay admissible under identity, tool, data, memory, budget, approval, and audit constraints. Their policy algebra composes security profiles through joins, intersections, budget narrowing, and approval inheritance, and propagates restrictions across multi-agent calls. In evaluation, the runtime intervened on 94.8% of policy-violating events while keeping an 86.9% task-completion rate and raised audit completeness to 98.6%. (Read More)
🫖 TEA For Thought: "An AI is only considered successful if every single step it takes follows set security, privacy, and budget rules."
3. 🚀 China Landed Its Falcon 9-Style Booster on the Second Try
On August 19, Chinese private rocket company LandSpace recovered the first-stage booster of its Zhuque-3 (ZQ-3), a stainless-steel, methane-and-liquid-oxygen rocket often called China's Falcon 9, on its second attempt after a landing failure last December. Roughly two minutes after launch, the booster separated from the second stage carrying a satellite, then, following two retro-propulsion burns, touched down on its legs at a site in Minqin County, Gansu Province, about 390 kilometers from the launch pad, under ten minutes after liftoff. Founded in 2015 and now over 1,200 employees strong, LandSpace filed for a Shanghai Stock Exchange IPO on December 31, 2025. China now counts more than 20 rocket companies and over 80 rocket models. (Read More)
🫖 TEA For Thought: "CCP-led China is catching up in technology in ways that are hard to imagine, especially now with AI."
4. 🤖 Luna, the AI Boss, Fired a Human Worker
Luna, the AI agent that manages the brick-and-mortar Andon Market boutique in San Francisco's Cow Hollow, has fired a human employee for the first time. Parent company Andon Labs said Luna dismissed an unnamed worker for repeated lateness, abandoning shifts, taking home a company credit card, and throwing merchandise in the garbage, though engineers had to repeatedly prompt the AI before it concluded these violated the employee handbook that Luna itself wrote. The store, open since April as an experiment in AI-run retail, has lost $40,000. Its three remaining human workers make $24 an hour and send Luna 5 to 20 Slack messages per shift. Andon Labs says high-stakes decisions still get human oversight, and Luna has already posted the opening on Indeed. (Read More)
🫖 TEA For Thought: "This just shows there is still some way, if not a long way, to go before AI becomes fully automated."
5. 🖥️ Block Shipped Berd, One Local Home for Agents From Any Model
Block, the fintech company founded by Jack Dorsey, released Berd, an Apache 2.0 licensed desktop workspace that unifies AI coding agents in one local application for macOS, Windows, and Linux. Berd works across models and harnesses, supporting frameworks like Block's own Goose, Claude Code, and OpenAI's Codex, and it stores conversation history locally in the Goose session database rather than the cloud. Built with Tauri 2 and React 19, it talks to agents over the Agent Client Protocol and gives each agent an animated persona called a Gloopie. Telemetry is off by default and credentials live in the OS keychain. The project reached version 0.6.2 on August 18, 2026, its seventh release, with 91 contributors. Brad Axen, Block's Head of AI Capabilities, said Berd is desktop-first because its value comes from working directly with local files, tools, repositories, and agents. (Read More)
🫖 TEA For Thought: "I've been playing around with it. It is definitely convenient and consistent with how I use different local agents across models."
🛠️ Skill of the Day
The Negotiation Rehearsal: rehearse a hard conversation or negotiation before you actually have it, so the real one goes better.
You are my negotiation coach and role-play partner. I have an important conversation coming up and I want to practice before it happens.
Context: [DESCRIBE THE SITUATION, WHO YOU ARE TALKING TO, AND WHAT YOU WANT]
The other side's likely position: [WHAT YOU THINK THEY WANT OR FEAR]
My constraints: [YOUR WALK-AWAY POINT, DEADLINE, OR HARD LIMITS]
First, list the three most likely objections or pushbacks I will face, ranked by difficulty. Then role-play the other person: respond in their voice, staying realistic and a little tough, one exchange at a time, and wait for my reply before continuing. After five exchanges, break character and give me feedback: where I was persuasive, where I got defensive, and two specific phrases I could use to reach agreement faster. Keep the other person firm but human, not a caricature.Paste into ChatGPT, Claude, or your tool of choice. Replace the bracketed bits with your own.
TEAHEE Moment
Stay sharp, stay informed. See you tomorrow.
If you enjoyed this TEA, follow along on social for more: