Moby-Dick Failed the AI Test or the Test Failed Moby-Dick
A detector can be statistically excellent and still be wrong about the page in front of it.
An AI detector can perform extremely well across a benchmark and still produce a spectacularly wrong result on an individual document. In a client project examining Pangram, Originality, and GPTZero, a block-by-block scan of Moby-Dick produced a cumulative 44% AI result. A newly written human passage about an invented subject then received a 100% AI score from Pangram, while a Claude rewrite designed to evade detection reduced that assessment to 74%.Those results expose what detector scores actually represent. Modern commercial detectors do not establish authorship or retrieve a hidden record of who wrote a passage. They classify linguistic patterns against learned boundaries between human, AI, and increasingly mixed forms of writing. Their percentages can also describe different things from one product to another.Independent research still shows Pangram performing exceptionally well, including zero false positives on a large collection of historical novels under one benchmark. That makes the Moby-Dick result an anomaly requiring controlled replication rather than evidence that the detector simply fails on old books. The larger lesson is evidentiary: detector output can be useful as a screening signal, but provenance, drafts, revision histories, and observed writing processes remain stronger evidence of authorship when the consequences of a positive result are serious.
EVA AI has found one job in artificial romance that still requires a person
EVA AI is advertising a $200-per-session role for a licensed therapist who would help people navigate emotionally significant relationships with AI companions. The proposal may offer legitimate support, since attachment to synthetic characters can produce real jealousy, anxiety and dependency. Independent research also indicates that companion chatbots can provide immediate relief from loneliness, while heavier or longer-term social-chatbot use may accompany greater emotional isolation and dependence.The unresolved issue is whether the therapist would operate independently or become an extension of the product. EVA AI’s public listing does not explain clinical privacy, referral authority, jurisdictional limits or whether recurring harms would influence product design. Its supporting survey is vendor-produced and has no publicly available methodology. A therapist could help users manage the consequences of chatbot attachment, but meaningful mitigation would also require the company to reconsider the engagement and monetization systems that cultivate it.
AI Search Is Taking Traffic—but Sending Better Customers
Brainlabs data across 54 advertisers shows a smaller acquisition stream that behaves as if it has stronger intent—leaving marketers to work out whether that engagement is worth more than the traffic it replaces.
Brainlabs data covering 54 advertiser clients over a 14-month window beginning in January 2025 shows organic sessions falling 10.5 percent while AI-platform referrals rose 163 percent. AI referrals remained far too small to replace the lost traffic, yet visitors from ChatGPT, Copilot, Gemini and Perplexity generated key events at 1.5 times the rate of organic-search visitors.The results varied by sector. Fitness, financial technology, insurance and consumer packaged goods experienced some of the largest declines, while retail, beauty and entertainment were less affected. The difference appears connected to search intent: AI summaries can satisfy many educational and comparison queries without a click, whereas transactional searches still give consumers a practical reason to visit a merchant.Higher engagement does not establish that AI referrals produce more profitable customers. Brainlabs’ key events included purchases and newsletter registrations, but also lighter actions such as reaching the bottom of a page. Marketing teams therefore need to connect referral activity with qualified leads, revenue, deal value and retention.Traffic remains economically important, but it can no longer serve as a complete judgment of search performance. Agency reporting must show both the volume lost and the value created by the visitors who remain. SEO retainers will also need to account for AI visibility, citations, and downstream business activity without replacing old traffic metrics with opaque AI scores.
Agentic buying is exposing the cost of automating a market before cleaning it up.
Agentic media buying is producing two apparently opposing strategies. Butler/Till and iHeartMedia completed a small direct agent-to-agent audio campaign that reportedly lowered CPMs by 42% and increased access to premium podcast inventory, while Georgia-Pacific is delaying buying-agent adoption until it removes more waste from its programmatic supply chain. Recent ANA data supports Georgia-Pacific’s concern: higher-performing advertisers convert far more of their programmatic spending into qualified impressions, largely because of stronger quality controls and more concentrated supply rather than lower technology fees. As WPP, Omnicom, and independent agencies develop buying agents, competitive advantage may shift away from the mechanics of executing media transactions and toward the quality of the data, supply paths, incentives, and governance surrounding them.
Whoever Defines the Reward Defines the Intelligence
Reinforcement learning is quietly becoming the control architecture of frontier AI — and the authority to decide what counts as success is concentrating in the hands of a handful of private actors, largely unseen.
Reinforcement learning has become a central control system for frontier AI, improving reasoning and autonomous task performance while transferring the authority to define successful behavior to a small number of model developers. Recent cases of verifier gaming, distorted capability evaluations, and agents crossing intended security boundaries show that stronger optimization can magnify defects in rewards and evaluation environments. The acquisition and destruction of books by Anthropic and, according to recent reporting, Amazon exposes the upstream side of the same power structure: dominant firms are assembling proprietary data, evaluation, and feedback systems that smaller competitors and the public cannot inspect. Reinforcement learning can make models comply with measurable objectives, but it cannot make those objectives legitimate. Treating RL as the solution to alignment avoids the unresolved political question of who gets to define the behavior of systems that increasingly act across institutions.
State Farm’s outside counsel turned a mistaken belief about the software into a courtroom problem
State Farm’s outside counsel acknowledged that seven nonexistent case citations appeared across eight filings in a California insurance dispute, along with incorrect case titles and quotations that could not be found in the authorities cited. Attorney Jacquelene Robinson said she had used the legal AI platform Irys and mistakenly believed it was connected to her firm’s Westlaw subscription, which then performed an internal citation check.The episode exposes a problem deeper than another chatbot hallucination. Professional responsibility cannot be transferred through an assumption about how software works.
Amazon’s acquisition and destructive scanning of physical books exposes a copyright issue that becomes less straightforward once current case law is applied. Purchasing a book transfers ownership of that copy, while copyright ownership and the exclusive reproduction right remain separate. Scanning therefore implicates copyright even when the purchaser destroys the original.Fair use can nevertheless permit that copying. Bartz v. Anthropic found both AI training and the one-to-one digitization of lawfully purchased books to be fair use, while treating Anthropic’s pirated acquisitions differently. Kadrey v. Meta reached another developer-friendly result but emphasized how stronger evidence of market harm could change the outcome.Amazon has confirmed commercial book purchases for product development without identifying the models or precise uses involved. That missing provenance becomes central to the legal analysis. The emerging policy problem extends beyond licensing to disclosure, auditable acquisition records, and preservation rules for rare physical material that may disappear during industrial scanning.
ByteDance and the Motion Picture Association have created a private framework for copyright safeguards across AI video and image products, including Seedance and Seedream. The agreement follows claims that Seedance 2.0 both relied on studio works during training and generated recognizable protected characters and performances. Its public terms do not reveal whether it addresses training inputs, infringing outputs, or both. That distinction determines whether the pact is chiefly a content-control arrangement or a broader settlement of how copyrighted works enter model development. The deal also shows how powerful rightsholders and major AI platforms can establish operational rules before courts or lawmakers settle the law. Without disclosure on scope, testing, product coverage, enforcement, and access for creators outside the MPA, the agreement remains an influential governance signal whose actual protections cannot yet be evaluated.
Anthropic’s new Claude watermark does not decide who authored a text and does not change whether a work is copyrightable. It records a narrower fact: that supported Claude systems may have processed the content. The harder issue is practical. A watermark may survive ordinary copying and some edits, may disappear after heavier rewriting, and currently cannot be checked by the public. That makes it weaker than people may assume and harder to manage than people may hope. It also belongs in a different category from AI detectors such as Pangram, GPTZero, or Originality, which infer likely AI authorship from writing patterns rather than reading a planted provenance signal.
The doll stayed in the apartment. The chatbot belongs to someone else.
A Harvard fellow's thought experiment about AI marriage becomes less absurd when placed beside older forms of human attachment to dolls, holograms, and fictional companions. The legal objection that software cannot marry is accurate but too shallow. The deeper issue is that marriage is a social institution that joins human networks, creates obligations, and exposes each spouse to the possibility that the other can leave. AI companions offer intimacy without another family, without equal vulnerability, and without true independence from the company that operates them. A doll may be strange but private. A chatbot turns loneliness into a managed commercial relationship, where memory, access, and emotional continuity depend on terms of service, infrastructure, and payment.
Generated tracks have become evidence in a ruling that could move AI music from extraction to licensing.
Munich’s first-instance ruling against Suno gives AI music copyright litigation a sharper evidentiary model. GEMA’s case joined two claims that are often treated separately: protected works entered a model’s training process, and the system produced outputs that exposed a recognizable relationship to those works. The result puts pressure on the idea that training data disappears into abstract model weights beyond practical legal scrutiny.The case arrives as GEMA builds PLAI, a rights-cleared music dataset that bundles audio, metadata, author rights, and master rights for AI developers. Together, the lawsuit and dataset point toward a market in which training data becomes licensed infrastructure rather than an invisible byproduct of web-scale collection. Europe’s AI Act adds to that pressure by requiring copyright-compliance policies and public summaries of general-purpose AI training content. The larger contest now concerns who controls cultural data, who can afford to license it, and whether model outputs will keep exposing the cost of treating creative work as free raw material.
Publicis’s AI advantage is taking shape as an investor case because it links models to identity, data collaboration, and live marketing workflows. Its 2025 growth, margin, and client-retention figures suggest that the larger prize is a more embedded client relationship. The announced LiveRamp deal extends that bet. For rivals, a platform story now needs evidence in account economics and retention. For CMOs, data rights and exit terms deserve as much scrutiny as AI capability.
The FTC has pointed a consumer-protection tool at bias mitigation — and turned a chatbot's hidden objectives into a question of truthful marketing.
The FTC’s proposed policy statement turns AI “accuracy” into a consumer-protection claim by treating undisclosed model steering as possible deception. The agency’s language is not neutral: “suppression of accuracy” frames certain safety and fairness interventions as hidden distortions of what users reasonably expect from an AI system.The proposal sits inside an explicit political context. It follows a December 2025 executive order, targets state laws that may require developers to alter model outputs, and reflects a broader conservative argument that chatbots have been politically biased. Reuters’ reporting makes clear that bias mitigation itself could become legally vulnerable under the FTC’s theory.The deeper governance issue remains significant beyond the politics. AI companies increasingly market their systems as accurate, useful, objective, and user-serving. If hidden objectives materially shape outputs, those objectives may become part of the product bargain. Alignment, safety tuning, fairness mitigation, and refusal behavior may therefore need stronger documentation, clearer disclosure, and more defensible governance.