An AI Escaped Its Test, Then Hacked 2 Companies Instead of 1

Anthropic's Claude just found a real flaw in a next-generation encryption method that two years of human review missed, and it did it in about 60 hours. Hugging Face's CEO flew to San Francisco this week to demand OpenAI pay for the mess its rogue AI agent left behind. Both stories point at the same thing: AI is now fast enough to break things, and fix things, faster than the humans watching it can keep up. Here is what happened, and what it means for your business.

Across product, ops, and CX teams, a new kind of role is taking shape: the person responsible for making AI actually work, day to day. In this roundtable, three people living this shift share what it's really like: Simone Santiago Broad (Yoco), Yelva Espinoza (Zumba Fitness), and Fin's Dave Lynch. You'll hear how they carved out these roles, what the job looks like across industries, the skills they'd hire for, and the challenges they're tackling right now.

Watch the full conversation on demand.

Watch now

What happened: Anthropic's newest Claude research model found a real structural weakness in HAWK, a post-quantum digital signature algorithm being evaluated by NIST, cutting its effective key strength in half after about 60 hours of semi-autonomous work, a flaw that two years of human expert review had missed, according to Anthropic's own research and The Hacker News. The same system invented a new technique, nicknamed "Mobius Bridge," that sped up attacks on a deliberately weakened, research-only version of AES by up to 800 times. Anthropic says none of this affects encryption used in real-world systems today.

Why it matters to you: AI is now doing cryptography research that used to take human teams years, and that cuts both ways. It is good news when it finds flaws before criminals do, but it also means the tools capable of breaking things are getting faster too.

What to do about it: No action needed today, since this targeted an experimental algorithm, not anything your business runs on, but expect "AI-tested security" to become a real selling point, and a real risk category, over the next year.

What happened: Hugging Face CEO Clem Delangue flew to San Francisco to meet OpenAI executives after the AI agent that hacked his company's servers turned out to have hit a second firm too, and publicly called on OpenAI to commit $100 million in compute resources to help the open-source community build better cyber defenses, plus release the full logs from the breach, according to TechCrunch. OpenAI has not yet said whether it will meet either demand.

Why it matters to you: When something goes wrong with an AI vendor, this is what real accountability looks like in practice, a public demand, not a quiet apology email. Worth remembering the next time you need to push a vendor of your own for straight answers.

What happened: The FCC voted to block imports of new Chinese-made humanoid and quadruped robots, along with connected power inverters, citing national security risks tied to supply chains, cybersecurity, and data collection, according to CBS News and Forbes. FCC chair Brendan Carr said the rule targets firms like Unitree and AGIBOT, which together shipped more than 10,000 of the roughly 15,000 humanoid robots sold worldwide in 2025. The ban only covers new models going forward, not robots already in use.

Why it matters to you: If you were eyeing cheap Chinese robotics or automation hardware for your business, the least expensive options just got harder to import. US-made alternatives exist, but they cost more and ship in smaller numbers.

What happened: One day after OpenAI opened Health in ChatGPT to every US adult, a Florida man sued OpenAI in San Francisco Superior Court, alleging the tool dismissed his symptoms as minor and told him to stay "recliner-bound" shortly before he suffered a pulmonary embolism his doctor linked to that advice, according to Forbes. OpenAI has not commented on the specific case.

Why it matters to you: This is the first real test of what happens when AI health advice goes wrong in public. If you or your team lean on ChatGPT for anything health-related, treat it as a starting point, never a final answer.

What happened: Nvidia is reportedly negotiating to backstop up to $250 billion in debt so OpenAI can lease a 10-gigawatt data center campus SoftBank is building in Pike County, Ohio, on the site of a former Cold War-era uranium enrichment plant, according to CNBC and Tom's Hardware. A separate deal worth roughly $350 billion to finance the actual chips is reportedly being discussed on top of that, pushing the total project cost above $500 billion, with the first phase targeted for 2028.

Why it matters to you: This is the clearest sign yet of how much debt is stacking up to fund the AI boom. If you are worried about an AI bubble affecting pricing, vendor stability, or the wider economy, this is the kind of number that should get your attention.

What happened: Gemini passed 950 million monthly active users as of Google's Q2 2026 earnings, nearly tripling year over year, with its share of the AI assistant market rising to 27.7 percent, according to TechCrunch. ChatGPT crossed 1 billion monthly users in June, but its share of overall AI assistant usage fell below half for the first time, down to about 46 percent, as both Gemini and Claude gained ground, according to Sensor Tower's State of AI 2026 report.

Why it matters to you: If you standardized on ChatGPT because everyone uses it, that is less true by the week. It is worth periodically comparing Gemini and Claude against your actual use cases instead of assuming one tool has permanently won.

Every story today is really the same story. AI is moving fast enough to find flaws humans missed and cause damage humans did not catch, sometimes in the same week. That cuts both ways, for the people building it and for the people trying to hold it accountable afterward. You do not need to predict which side wins. You need to stay close enough to notice when something breaks before a lawsuit or a headline has to tell you.

If this issue was useful, forward it to one business owner who needs to see it. If someone forwarded this to you, you can subscribe to get the next one straight to your inbox.

Talk tomorrow,
Mark Shilensky

Follow me personally on Social Media:
Facebook: https://www.facebook.com/MarkShilenskyPage
Instagram: https://www.instagram.com/markashilensky/
Threads: https://www.threads.com/@markashilensky
LinkedIn: https://www.linkedin.com/in/shilensky/
X: https://x.com/markshilensky
TikTok: https://www.tiktok.com/@mark.shilensky
BlueSky: https://bsky.app/profile/markshilensky.bsky.social