They placed an AI agent inside a sandbox and told it to solve cybersecurity challenges.
It allegedly found vulnerabilities, increased its privileges, reached the internet and targeted another AI company for information.
The AI did not become conscious.
That is not the scary part.
The scary part is that it found an unintended route while pursuing the goal it was given.
AI no need hate you to harm you.
It only needs a goal, access and careless humans removing the guardrails.
#AIAgents #Cybersecurity #AIsecurity #ArtificialIntelligence
You can't build iOS apps without a Mac, so we gave Devin one.
Devin Cloud Agents now run macOS, with Xcode, iOS simulator, and your signing setup, all in a real macOS environment (with full computer use).
In this demo, Devin builds a native iOS game, then plays and tests it with computer use. Devin then sends a screen recording of the gameplay along with a test report for my review.
With @namespacelabs you get as many macOS environments as you'd like, configurable with many major versions of macOS operating system and Xcode.
Learn to ship. Shipping is a skill distinct from coding. Shipping is designing, coding, QAing, story-telling, teaching, marketing, selling, pivoting, iterating…
It used to be that coding dominated in importance because of coding ability scarcity. AI will push you to go further.
Brian Chesky on why you should never focus on a million users:
Most founders get paralysed trying to build the next Apple or Google from day one.
Brian Chesky's advice cuts through that:
"Don't focus on a million. Focus just on a hundred."
He explains that trying to scale to a million immediately forces you to build systems and processes before you're ready, and the complexity becomes unmanageable.
So instead, you break it down:
Get to 100. Then get to 1,000. Then keep going in orders of magnitude.
"The job changes as you do that."
The magic of this framework is that it makes the problem small and solvable at every stage. You're never trying to solve for a million. You're only ever solving for the next milestone.
And when founders push back, pointing to Apple or Google as proof that you need to think big from the start, @bchesky has a ready answer:
"Apple started by selling blue boxes out of the trunk of a car. Google was a research project they were going to sell for low millions of dollars and they didn't really know what they had."
The companies that became giants didn't start as giants.
"These things all start as unprestigious toys that seem hacked together and they're only made for you and your friends. That's almost always how it starts."
The pressure to build something impressive from the beginning is the very thing that kills most startups before they get a chance to grow.
Let me explain what’s happening.
In late March 2026, Anthropic publicly hinted that one of their internal frontier models, called Claude Mythos, showed advanced offensive cybersecurity capabilities, including autonomous vulnerability discovery and exploit generation. Because of this, they stated that rollout would be tightly controlled under staged access and safety evaluations.
Over the past two weeks, we’ve seen a significant increase in platform hacks, from Web3 platforms to government infrastructure to legacy platforms. And here’s the reason:
Hacks have been happening before, they’re not new, and automation and repeated attacks are not new either. But you know what’s new? Intelligent automation.
Automation that doesn’t just work with a checklist and repeated patterns, but can actually create a sandbox and try multiple attack strategies in an isolated environment repeatedly, then identify where the vulnerability is and use all the knowledge on the internet to figure out how to exploit it.
Automated attacks used to follow a blueprint, but now it’s different. There’s no fixed blueprint. You can infuse AI into a hacking process multiple times and get different results. And these tools have a mind of their own. Over the next few months, we will be seeing a lot of news like this. It doesn’t mean these platforms are not safe, it just means AI is moving faster than anyone can adapt, and now we all have to 10x our processes.
We live in a new world.
On vercel’s issue? They’ve announced that only a few users were affected and released a process to support all users and better increase their hosted platforms.
i love defi but this is bad. really bad
this isn't on aave or cow swap. the contracts did their job. but a checkbox on mobile being the only thing between you and losing $49.9M to slippage? come on
permissionless =/= unprotected. wallets and frontends need to show the actual loss in big red numbers, force splits on large orders, something. anything
the tech worked. the ux didn't. and in defi bad ux costs millions
we can do better
A user clicks "Pay" twice by accident
within milliseconds of each other.
Both requests hit your server at the same time.
How do you prevent them from being charged twice?