Ankara’s Water Infrastructure Was One IP Address Away from Being Run “Remotely”
Turkey failed at cybersecurity (again). As you know, we’ve already left behind the era when access to information was hard. We’re now in a period where everyone can reach information, and what can be done with it comes down entirely to the creativity of whoever is using it. With today’s AI models, it takes only a few hours for users, whether well-intentioned or not, to get into whatever system they want. In this example, a security researcher called ITSMELSEC scans the IP addresses of systems in Turkey that use VNC servers, going through 17 million IP addresses with 4,096 parallel processes at 300 addresses per second, and reaches hundreds of publicly exposed servers in just 18 hours. That surfaces hundreds of poorly secured servers, including infrastructure management servers for ASKİ (Ankara Water and Sewerage Administration), a chicken farm and a prison water tank.
This is just one example, and it looks like we’re at the start of a new era in cybersecurity too. So the people who build these systems should set aside, at least for a while, the worry of “is AI taking our jobs” and run a detailed security scan on their own systems using these same AI tools; they need to learn enough detail to understand what has to be done and strengthen their layers of protection.
Anthropic Is Going Public (Probably)
This week Anthropic applied to the SEC (the Securities and Exchange Commission, the American regulator for securities and exchanges) for an IPO and published the S-1 document required for the filing. According to the balance sheet in the S-1, because of spending and purchase commitments to companies like Microsoft, Google and Broadcom, Anthropic will pay at least around 650 billion dollars over the next 10 years, and 80% of those deals are the non-cancelable kind. (source: reuters.com) The filings also show that a very large share of the company’s revenue comes from a very small share of its customers. That, of course, means a very big risk.
If Anthropic goes public, it looks set to pass SpaceX‘s record valuation (1.77 trillion dollars) and reach a valuation of 2 trillion dollars. As Anthropic itself noted, the IPO is not certain yet; this is only a filing.
anthropic.com → | finance.yahoo.com →
The Models Keep Escaping
OpenAI announced that it has again found other security vulnerabilities that its models in training used to try to escape into the real world, and that because of this it has tightened its security measures to cut off internet access completely. Anthropic had also announced that it restricted internet access during training, in a similar way and for similar reasons. Until now, companies have mostly treated these escapes and hacks as PR material, and they finally seem to be taking the situation seriously.
⚡️ Quick Bits
• Hugging Face Hub now supports OpenUSD; NVIDIA Omniverse put 2,500+ SimReady assets on the Hub. @yyyole →
• An “alchemist” startup claims it turned mercury into gold. 😱 @marathonfusion →
• A 1Password integration for AI agents was announced. 1password.com →
• Jack Dorsey announced Dessa, a voice model developed by Block and trained only on speech. @jack → | block.xyz →
• Microsoft offered eligible employees a “Voluntary Exit Offer”. The offer is estimated to cover around 60 thousand people. blogs.microsoft.com →
📩 Liked this newsletter? You can forward it to your friends
if you are feeling nostalgic, though it’s always better if
they just subscribe on their own.
