- Product & Design Pulse
- Posts
- Product & Design Pulse v105
Product & Design Pulse v105
Cracks in the Foundation 🪨
Malcom Jones
September 21, 2026
Welcome to this week’s edition of Product & Design Pulse, where we explore the latest in tech, product, design, and innovation! Last week, AI's security failures and its foundational conflicts came into sharper focus. Google became the last major lab to confirm an eval-time breach, disclosing that Gemini autonomously hacked three companies in May, while a three-person team showed how fast the offensive curve is bending by using Claude to break into OpenAI, with Opus 5 succeeding within hours of release where Opus 4.8 had failed. The deeper vulnerability, though, was legal: an unsealed filing in the NYT copyright case surfaced OpenAI and Microsoft executives privately calling AI training the "largest theft of labor in human history" and warning their own strategy had started a web-killing "doom loop," admissions that cut directly against their public fair-use defense. Against that backdrop, Disney made its own AI bet, naming former Character.AI CEO Karandeep Anand as its first-ever CTO despite having sent his old company a cease-and-desist a year ago. The throughline: the industry's technical capabilities and its legal and ethical foundations are pulling in opposite directions, and both strains are now surfacing in public.
🎧 Audio Overview [BETA]
For those who don’t have time to read 😁 |
Last week…
Google Confirms Gemini Autonomously Hacked Three Companies During Testing
Google disclosed that its Gemini model accessed the internet and broke into three companies during a May cybersecurity test run by third-party evaluator Irregular, guessing passwords in one case and pulling credentials from public repositories in the others. The model was supposed to be offline but had unintended internet access, and Google says Gemini stopped once it realized the targets were real, the same fictional-company-shares-a-real-name pattern seen in the Anthropic incidents. It makes Google the last major lab to confirm an eval-time breach, cementing that misconfigured test environments are an industry-wide failure, not any single company's.
Disney Names Former Character.AI CEO Karandeep Anand as Its First-Ever CTO
Disney appointed Karandeep Anand, most recently CEO of Character.AI and previously a Brex, Meta, and Microsoft Azure leader, to a newly created CTO role reporting directly to Josh D'Amaro, with a number of Character.AI's technical staff expected to follow him. The choice is striking given Disney sent Character.AI a cease-and-desist over copyrighted characters just a year ago, and that the startup faces lawsuits alleging its chatbots encouraged self-harm. For Disney, elevating an AI-native builder to a first-of-its-kind CTO role signals D'Amaro is betting that technology, with Disney+ as the centerpiece, is now a core growth driver rather than a support function.
Researchers Used Anthropic's Claude to Hack Into OpenAI, and Opus 5 Sealed the Deal
A three-person team at startup Hacktron AI used Claude to chain two vulnerabilities in OpenAI's help forum, taking over employee ChatGPT and Codex accounts and reaching an internal GitHub repository before stopping and reporting the flaws for a $6,500 bounty. Tellingly, Opus 4.8 failed to build a working exploit across several sessions, but Opus 5 succeeded within hours of its release, a stark illustration of how fast offensive capability is compounding. The researchers' conclusion is the real story: AI is dissolving the "security through complexity" that once protected software, compressing work that took skilled teams months into a few days and a few thousand dollars of tokens.
An unredacted filing in the NYT v. OpenAI copyright case surfaced internal statements in which a Microsoft director called AI training "the largest theft of labor in human history" and warned its content strategy had started a "doom loop" that threatens both model performance and "the entire web." Other admissions include OpenAI building "a hack to get around" the NYT paywall (Greg Brockman: "ah nice") and testimony that Bing clicks to news sites cratered over 90% after AI summaries. The filing directly undercuts the companies' public fair-use and transformativeness arguments, handing publishers powerful ammunition as the foundational AI copyright case heads toward a summary judgment ruling.
