Zendoric

AI analysis and perspective, every day.

Updated: September 23, 2026

🎙️ Listen to today's podcast (in Spanish)

🔥 In focus today

The ad becomes an agent: OpenAI tests sponsored chatbots in ChatGPT, and labeling is the only firewall

🕒 Published on Zendoric: September 23, 2026 · 02:26

OpenAI is testing Sponsored Agents — business-funded chatbots that pick up the conversation when you click an ad inside ChatGPT. The company says these sessions stay separate from ChatGPT's own answers and are clearly labeled. That separation is not a detail: it is the entire governance question of AI-native advertising, and it arrives with a $1 billion ad business already behind it.

Read the analysis🎬 Video

Sanders wants 20 years in prison for building superintelligence — but nobody can define the crime

🕒 Published on Zendoric: September 23, 2026 · 02:26

Bernie Sanders and Greg Casar have introduced a bill that would criminalize developing artificial superintelligence with penalties borrowed from nuclear weapons law, and freeze frontier AI until a new federal agency exists to license it. The instinct is defensible: safety rules written by the companies being regulated aren't rules. The execution isn't: you cannot prosecute a category that has no measurable threshold, and the one genuinely good idea in the bill is buried under the one that won't survive contact with a courtroom.

Read the analysis🎬 Video

Claude flagged a novel enzyme system in 21 hours — but the Crispr comparison rests on a function nobody knows yet

🕒 Published on Zendoric: September 23, 2026 · 02:26

Anthropic says nearly 950 Claude agents burned 210 million tokens over 21 hours to surface a previously uncharacterized enzyme system in bacteria-infecting viruses, the first result from its new wet lab. The company is comparing it to Crispr while admitting it doesn't yet know what the enzyme does. The scale of the search is genuinely new; the certification is missing.

Read the analysis🎬 Video

One Hacker, Three AI Vendors, $8,000: Agentic Attacks Just Lost Their Cost Ceiling

🕒 Published on Zendoric: September 23, 2026 · 02:26

A five-day campaign against 100+ companies, run on agents built from Anthropic, DeepSeek and Moonshot models, reportedly cost the attacker $8,000. The headline is not that AI can hack — it's the unit price: under $80 per targeted company. And the fact that one operator mixed three vendors across two jurisdictions tells you the model layer is not a chokepoint.

Read the analysis🎬 Video

📰 More AI analysis

Anthropic puts numbers on itself: Claude now "leads" 26% of the R&D that creates the next Claude

🕒 Published on Zendoric: September 23, 2026 · 02:26

Anthropic has published three unprecedented metrics — R&D automation, agent supervision and compute allocation — to measure the pace of AI's self-acceleration. The figure that carries the most weight: Claude already "leads" 26% of the research building its successors, though the company is measuring itself.