Good morning. There are two kinds of people in politics. The ones knocking on doors with clipboards, and the ones writing checks so large they come with their own zip code. Guess which group AI companies are joining (hint: it’s not the door knockers). AI labs are pouring money into the midterms to shape the rules for AI, data centers, copyright, labor, and safety. Groups linked to OpenAI and Anthropic have already spent a combined $37 million on campaigns. The weird part is that they are funding both sides. That’s not ideology. It’s insurance.
Gartner says AI coding will soon cost more than developers. AI coding companies are ditching flat monthly subscriptions for token based pricing. Every time an AI agent thinks, retries, loops, or pulls in more context, the bill gets bigger. Gartner says many companies still assume more tokens mean more productivity, even though it sees no evidence the two are directly linked. That helps explain why monthly AI coding bills are climbing from a few hundred dollars to thousands, and sometimes tens of thousands, per developer. Gartner predicts AI coding costs will surpass the average developer’s salary by 2028 if companies keep treating tokens like free candy. AI vendors have every reason to celebrate tokenmaxxing. They’re the ones selling the tokens. (Gartner)
China claims to have its own Mythos. Chinese cybersecurity giant 360 Security says it built an AI bug hunter to rival Anthropic’s model. There are no public benchmarks yet, and Reuters could not verify the numbers, but 360 claims Tulongfeng found 3,432 software flaws, with Chinese authorities confirming 105. 360’s founder framed Mythos as an existential threat, arguing the US cannot be the only country with AI that scans for weaknesses at scale. He also admitted Chinese models still trail US models, but said 360 can close the gap by pairing AI with its own security data and automation. Sounds like agent swarms trained to find zero days. Pray this thing never hits Hugging Face, the cybersecurity world isn’t ready. (Reuters)
Everyone knows LLMs hallucinate (usually with confidence). You.com’s guide shows how AI grounding connects models to trusted data so teams get more accurate answers and fewer fiction projects.
This playbook covers:
- Why grounding beats RAG alone
- How to build audit trails
- Open vs. closed platforms, and what it means for your next model switch
👀 closer look
The AI Cold War is starting to worry researchers on both sides. The US and China have spent years treating AI like a race someone has to win. Washington imposed chip controls to slow China down, while Chinese labs pushed powerful open models worldwide. Now researchers in both countries warn that more capable agentic AI could make cyberattacks faster, cheaper, and harder to contain. MIT’s Stephen Casper says the US and China should work together before AI gets its own “Chernobyl moment.” Cooperation sounds nice until both countries remember they’re still trying to beat each other. Regardless of who wins the AI Cold War, everyone loses if AI is developed recklessly. (Wired)
A physicist just threw cold water on Microsoft’s quantum breakthrough. Microsoft spent more than 20 years chasing Majorana, the strange particle it says could make quantum computers easier to scale. The problem is that Microsoft still hasn’t convinced scientists it showed up in its chips. UK physicist Henry Legg says Microsoft’s verification tool has coding errors and isn’t accurate enough to support the claim. Microsoft stands by its work and says it has shared data with DARPA. But outside scientists still can’t fully verify the breakthrough. For now, the strongest signal coming from Microsoft’s quantum chip may be the marketing department. (BBC)
AI companies are practicing medicine. They call their tools support software, wellness apps, or patient education because those labels keep them outside the FDA’s medical device lane. That loophole matters now that hospitals are giving doctors AI clinical reasoning tools. According to a 2026 AMA survey, 80% of doctors already use AI at work. Even bigger, a recent study by Harvard and Stanford researchers tested ChatGPT on real patient cases and the bot outdiagnosed hundreds of physicians. This doesn’t mean AI is ready to take over medicine. It means the line between “support tool” and “doctor” is getting blurry fast. These apps now help doctors think through cases, draft patient messages, and read lab results, while every company involved insists this is definitely not medical advice. Amazing how much medicine you can practice when you claim you aren’t. (The Atlantic, Science)
fun stats
👨💻 55%. New hires at Big Tech who were engineers in 2025, up from 46% in 2019. AI hasn’t replaced engineering jobs. It’s made them even more valuable.
🌶️ 9 months. Time it took OpenAI to go from zero to Jalapeño, its first custom AI inference chip built with Broadcom.
🛸 5,000. Small military drones France is ordering from defense unicorn Harmattan AI as Europe races to ramp up drone production.
🦾 $2.5 billion. Valuation Agility Robotics is targeting as it goes public. The humanoid startup has secured $300 million in orders for Digit, its bipedal robot already working in Toyota factories.