Hey,
Let’s unpack what’s new…
TOP 3
Anthropic says its restricted Mythos AI model has already uncovered more than 10,000 serious software vulnerabilities through its new cybersecurity initiative, Project Glasswing. More than 1,700 were confirmed as real flaws, including critical bugs capable of compromising major systems.
That sounds impressive until you realise the uncomfortable part: AI is now getting extremely good at breaking software faster than companies can repair it.
Anthropic admitted the imbalance directly. Finding flaws is becoming easier than fixing them. Which means the same frontier models helping defend critical infrastructure could also quietly make cybersecurity far harder over time.
Supported by rocket.new
Most competitive tools surface raw data and leave the interpretation to you. Rocket Intelligence monitors companies across 9 signal pillars, hiring, pricing, ads, tech stack, exec moves, reviews and tells you what it means for your role specifically. A pricing page change plus a hiring surge in enterprise sales means something. Rocket connects the dots so you don’t have to.
TOP 3
Google Cloud COO Francis de Souza says companies need to stop treating AI security like something they can “bolt on later,” warning that shadow AI, unsecured agents, and forgotten internal systems are creating entirely new risks inside enterprises.
The weird part is that even Google appears to be learning this in real time.
Recent reports found multiple Google Cloud developers were hit with massive unexpected Gemini API bills after old API keys quietly gained access to AI services they never intended to enable. In some cases, attackers reportedly continued using compromised keys even after developers deleted them because revocation delays took time to fully spread across Google’s systems.
The bigger issue is that AI agents are starting to move across workflows, documents, apps, and internal infrastructure much faster than companies can properly secure them. Right now, even the platforms building the AI stack are stress-testing their own systems in public.
TOP 3
OpenAI says one of its reasoning models solved a major geometry problem that mathematicians had struggled with for nearly 80 years. Researchers involved say the result would likely deserve publication in a top mathematics journal even if a human had discovered it.
What makes this different is that the model didn’t just remix existing proofs or solve a benchmark puzzle. It explored a completely different direction that many mathematicians had mostly ignored for decades because it looked too tedious and unlikely to work.
That’s what has researchers paying attention. AI is starting to move beyond accelerating human work and occasionally producing genuinely useful discoveries on its own.
Humans still verified and refined the proof, but several mathematicians involved admitted the model explored paths people probably never would have seriously pursued themselves.
In other developments
Pope Leo XIV is releasing a major AI-focused encyclical alongside Anthropic co-founder Chris Olah, positioning artificial intelligence as one of the defining moral and societal challenges of the modern era. The Vatican’s growing involvement in AI debates signals increasing concern around human dignity, labor, and the role of AI in shaping human identity.
Ferrari is using IBM’s AI tools to turn casual Formula One viewers into year-round superfans through personalised content, AI-generated race insights, and interactive fan experiences inside its app. The partnership shows how sports teams are increasingly using AI not just for performance analytics, but to deepen fan engagement and loyalty.
Advertise with us
Looking to promote your company, product, service, or event to 100,000+ AI leaders?