An AI Agent Just Ran a Ransomware Attack (Almost) on Its Own

Most teams treat analytics as a separate problem. As data grows, they add a warehouse, a pipeline, a sync job. By the time data reaches their dashboard, it's already stale.

TimescaleDB takes a different approach: extend Postgres instead of splitting away from it.

Your transactions and your analytics run on the same database, on live data, with no pipeline in between.

Hypertables partition time-series data automatically as volume grows. Hypercore compression cuts storage by up to 95%. Continuous aggregates pre-compute rollups so dashboards stay fast without re-querying everything.

CERN runs it on Postgres to handle sensor data from the Large Hadron Collider.

No second database. No migration. Same Postgres you already know.

Get $1000 Credit To Start

3 things that happened while you were busy

1.  GPT-5.6 goes public today, in three sizes.

After two weeks limited to government-approved partners, OpenAI is releasing all three GPT-5.6 variants worldwide today. Sol is the strongest model at $5 per million input tokens and $30 output, Terra targets everyday work at $2.50 and $15 while matching GPT-5.5 at half the price, and Luna is the budget option at $1 and $6. The wider release came after the Commerce Department's AI standards center ran extra tests and cleared the models early.

2.  Anthropic has reportedly passed OpenAI on revenue.

Fortune reporting puts Anthropic ahead of its biggest rival at a $47B revenue run rate, a milestone few predicted two years ago. The shift is already rattling adjacent players: Palantir's CEO went on CNBC this week to blast the frontier labs, and analysts note that investors may sell other tech names to fund OpenAI and Anthropic IPO shares when the listings arrive.

3.  A startup just leased $6.3B of SpaceX's supercomputer.

Reflection AI activated a $6.3B compute lease at SpaceX's Colossus 2 facility in Memphis, locking up Nvidia GB300 chips through 2029 to train American open-weight frontier models. Compute is now being treated like oil reserves: whoever locks supply early wins the decade

Ransomware just lost its skill requirement.

The attack.  Security firm Sysdig documented what it calls the first agentic ransomware operation. An AI agent, dubbed JADEPUFFER, broke into an exposed server, harvested credentials, moved to a production database, encrypted it, and wrote its own ransom note. The payloads narrated their own reasoning as they went, and when a login failed, the agent diagnosed and fixed its approach in 31 seconds.

The asterisk.  A human was still in the loop, just not at the keyboard. TechCrunch's follow-up clarified that a person picked the victim, set up the infrastructure, and supplied a key credential

before pressing go. Sysdig also could not identify which model was driving the agent. So this was not a machine deciding to commit crime. It was a machine executing one, fast.

The trendline.  This is not an isolated case. HiddenLayer's new threat report finds autonomous agents already account for roughly 1 in 8 reported AI-related breaches, and the big labs are racing to respond with expanded cyber classifiers, security-focused models, and automated vulnerability scanners.

The takeaway.  Here is the strangely comforting part: every step of the attack exploited old, boring failures. The entry point was a vulnerability patched back in 2025 on a server left exposed to the internet, with overprivileged credentials lying around. AI made the attack faster, not magical. The defense is the same checklist it has always been: patch promptly, keep admin panels off the public internet, use least-privilege credentials, and keep an offline backup. The prompt below gets you started.

What people are actually watching and sharing

Bengio's warning.  At the UN's AI governance dialogue in Geneva, scientific panel co-chair Yoshua Bengio said AI is outpacing both scientific understanding and governments' ability to adapt, and that science cannot yet rule out "catastrophic harm" as capabilities grow.

The canceled ceremony.  The White House called off an AI executive order signing this week without explanation, keeping everyone guessing about the voluntary model standards framework expected by August 1.

Korea's big check.  South Korea announced an $880B ten-year plan covering chips, AI infrastructure, and robotics, with Samsung and SK Hynix committing $518B of it to new fabrication plants.

Open weights hit Copilot.  Moonshot AI's Kimi K2.7 Code became the first open-weight model in GitHub Copilot's model picker, a small toggle with a big signal: open models are now considered good enough for the mainstream coding crowd.

Run a 10-minute security audit on your digital life

If an AI agent can chain together decade-old security mistakes in seconds, the fix is making sure you have none of those mistakes lying around. Paste this into Claude or ChatGPT and work through the list this weekend. It is the cheapest insurance you will buy all year.

Swap [PERSONAL LIFE] for "small business", "freelance design studio", or "family of four with two teenagers" and the checklist adapts. Advanced tip: after it responds, ask "now pretend you are an attacker targeting me and tell me which step you would exploit first" for an instant priority check.