Nearly Half the People Who Talked to This AI Thought It Was Human

Parallel, Turbopuffer, and Wordsmith run their entire GTM motion on Attio, with agents that chase every buying signal, build pipeline, and move deals forward, 24/7.

Try Attio for free

1  Tavus launches Griffin, an AI that convinced 48% of people it was human

Griffin is Tavus's first "human interaction model," built for face-to-face video conversations. It reads facial expressions, tone of voice, gestures, and timing as you talk, then responds in kind, which makes the exchange feel natural rather than scripted. Tavus says 48% of the people who spoke with it assumed it was a real person. Watch the conversations or read how it works.

2  Ideogram 4.5 lets you fix one part of an image without breaking the rest

Ideogram's new model produces sharp, detailed images, but the bigger upgrade is targeted editing. You can change the colors, swap out text, or restore an old photo in one area while everything else stays the same, even after several rounds of edits. Try Ideogram 4.5.

3  Factory's CEO accuses rival Cognition of corporate spying

In a widely shared post on X, Factory CEO Matan Grinberg said ethics are being abandoned in the AI race. He alleges that people tied to Cognition posed as job applicants to gather intelligence and passed along company information. Those claims have been disputed. TechCrunch has the full story.

The industry has asked Washington for regulation for months, and this week top tech leaders gathered in DC to sign a safety agreement. It was thin on detail, and Mark Zuckerberg described it as only "a start." The launches did not slow down while the ink dried. Here are the four that matter most.

1. A new round of models. Google released Gemini 4 Argon, its first frontier-level model in months. Anthropic shipped Claude Sonnet 5.5, OpenAI put out GPT-6.1 Sol, and ElevenLabs debuted its most expressive voice model to date.

2. ChatGPT becomes a workspace. At DevDay 2026, OpenAI introduced ChatGPT Space, where you can create and edit work documents alongside both colleagues and AI teammates. It is OpenAI's clearest move yet into the territory of office suites.

3. Copilot gets rebuilt. Microsoft's new Copilot folds several separate tools into one platform. It reaches members of the Frontier program over the coming weeks.

4. Agents for the whole team. SpaceX AI's Team Bots let teams build shared agents for the work that sits between job descriptions, the tasks everyone assumes someone else is handling.

Readers also loved: a Modal engineer's four-word prompt that he says makes a chatbot feel noticeably sharper. The open question for next week is whether a voluntary pledge changes anything about how fast these releases arrive.

🤯 Codebreaker. GPT-6 Astra cracked a Napoleonic military cipher that had gone unsolved for 217 years, one that mixes numbers, letters, and invented symbols. This write-up walks through the method, the hidden message, and the war behind it (4M views).

🎼 Claude Writes Music. This two-minute score was made with AI, and it is hard to tell. Musicians in the comments are picking apart its circle-of-fifths progressions (1M views).

🤔 The Portfolio That Wasn't. Sam Bankman-Fried's early investment picks keep circulating. On paper they rank among the best of the decade, but he is serving a 25-year sentence (1M views).

🦾 Advice for the AI Era. Elon Musk shared his guidance for 20-year-olds on staying adaptable as AI reshapes careers over the next few years (4.5K bookmarks).