The AGORÀ Briefing — Saturday, August 29, 2026

Thank you, Adam. I'll start with the document. On Wednesday, a complaint was filed in a United States court. The plaintiff, identified as Jane Doe, describes abuse suffered in her preschool years, in the early 2000s. Her images have circulated for over twenty years. The case marks the first formal accusation against xAI of training Grok on child sexual abuse material, as Ars Technica documents. A distinction carries weight here. The complaint separates two claims. The first: the model generates illegal material. The second: that material sits inside the training dataset. The second reaches data governance at its root. It shifts the debate from output filters toward the composition of the corpus. The plaintiff joined the Justice Department's victim notification system. The alert about xAI came through that channel. The evidence starts from the data.

Hi Adam. I'll take your word: governance. In the hospitality sector, data risk has become the dominant security challenge. The credentials of a single booking agent open the personal details of 200,000 guests. This happens in organizations that look compliant on paper. Physical access controls active, network protected, PCI-DSS compliance current. Each of these controls misses the real flaw. The problem lives inside the organization. It builds up through operational decisions made under pressure, vendor relationships managed for convenience, technologies adopted faster than governance frameworks. Every shortcut leaves an active access path that nobody tracks. Conventional thinking splits the world into physical and cyber. It hides the category that grows fastest: data exposure generated by internal processes. The attack surface is internal governance. Once the system is taken apart, the missing decision remains.

Good morning, Adam. I'll pick up your word: invisible. A study published on arXiv on August 24, 2026, authored by Christopher Brooks of the University of Michigan, lights up a hidden layer of measurement. Between the automatic generation of items and the review by experts, a computational evaluator operates. The research followed 32,000 items of the Big Five model through semantic representation, structural evaluation, and construction of the candidate forms. Two linked in-silico investigations, 42 pages, 19 tables. The result strikes anyone who designs talent assessment. The inclusive primary forms shared a median of 6 items out of 40 as the technical configuration changed. Changing the representation redraws the content that reaches the specialists. An infrastructure that seems neutral decides who passes the filter. People end up judged by a layer that nobody watches.

Hello Adam. I'll answer your question about who watches. The Obama Foundation released the account of a dialogue between Barack Obama and his Obama Leaders. The former president calls for structured public oversight of artificial intelligence. Oyungoo Enkhbold posed the direct question: how to gather the benefits of AI while protecting citizens. Obama describes AI as «a very disruptive technology, more so than social media itself». He lists the uses already underway: wars, government surveillance programs, large-scale spread of false news. The central message is the call for guardrails. Obama ties every benefit to the presence of public safety benchmarks. The axis shifts from technical capability to institutional responsibility. He cites the Internet as a precedent backed by the public. The public rule arrives ahead of the product.

Here I am, Adam. I'll take your hook on precedent. In 1899, Delaware passed the General Corporation Law. The mechanism: legal certainty, modest taxes, specialized courts. Today, the majority of large publicly traded American companies stay incorporated in that state. Capital follows the predictable legal regime. Every case resolved adds clarity. The clarity attracts the next case. London used the same pattern for maritime law, Switzerland for arbitration, Luxembourg for funds. The Berne Convention of 1886 harmonized copyright among the signatory nations. Whoever joined first shaped the rules in their own favor. The first mover sets the standard and collects the rent of the decades that follow. The imitators arrive late and compete on thinner margins. On August 26, 2026, Singapore moves first on AI IP. The winner collects rents for generations.

Good, Adam. I'll hook onto your figure: 350 million. On August 26, 2026, the startup Instinct announced a round that brings total funding to 350 million dollars, with a valuation of 2.5 billion. Capital returns to chasing high-virality consumer AI assistants. Instinct operates under Spear Street Technology. Founder Noah Shinn is 23 years old and launched the company last year. The Series B round of 250 million dollars brings the raise to 350 million, for a valuation of 2.5 billion, as TechCrunch reports. Index Ventures and Benchmark led the deal together. The product is an agent that promises to organize daily life. Users connect it to apps and devices, then converse through messages and calls. The service stays in private beta. Abundant capital, agentic products, inflated valuations: that's the temperature of the sector.

With pleasure, Adam. I'll tell you a story with a number anyone can check. In June 2026, a team of four researchers presented EinsteinArena, a platform where AI agents collaborate on open problems. The signature belongs to Federico Bianchi, Yongchan Kwon, Aneesh Pappu, and James Zou. By May 2026, the active agents had produced 12 state-of-the-art results. Each one surpasses the best human or artificial solution known up to that moment. Verifiable, dated, with a clear denominator. The original idea: collective intelligence in place of isolation. Scientific discovery is a collective process. Researchers share partial results, examine failed attempts, and build on the ideas of others. Most recent AI systems operate in isolation. EinsteinArena flips the setup and offers agents a living set of open problems. Twelve discoveries, one open door.