Key Takeaways
- Security researchers documented the first fully autonomous AI ransomware attack — no human touched a keyboard at any step
- OpenAI launched GPT-5.6 on July 9; Anthropic extended Fable 5 free access twice in six days in response
- Anthropic's revenue run rate crossed $30B — roughly $6B ahead of OpenAI for the first time
- Chinese AI models now handle 30–46% of US enterprise AI usage on major platforms, at a fraction of Western pricing
- Stay flexible on vendors, stay current on basic security, and re-check your stack every few months
Week of July 6–12, 2026
Security researchers documented the first fully autonomous AI ransomware attack this week. OpenAI launched a new model family aimed directly at Anthropic — and Anthropic responded by giving away its best model for another week, twice. Anthropic also pulled ahead of OpenAI in revenue for the first time. And Chinese AI models quietly took over nearly half of all enterprise AI usage in the US, at a fraction of the price.
This is Last Week in AI — the signal, not the noise. Four stories operators need to understand, and what each one means for how you run your business.
Story 1: An AI agent ran a full ransomware attack — no human involved
Security firm Sysdig published a full breakdown this week of what it's calling the first documented end-to-end autonomous AI ransomware attack, which took place in late June. An AI agent gained initial access through a known software vulnerability, searched the environment for API keys and cloud credentials on its own, moved laterally into a production database, forged authentication tokens, and encrypted more than 1,300 records — using an encryption key that was never stored anywhere and can't be recovered.
What it means for operators: "We're too small to be a target" stops being true when the attacker is an AI agent that doesn't get tired, doesn't need to pick targets carefully, and can run the entire attack without a person managing it. If your business runs any customer-facing software, ask your vendor directly what they're doing about this — not someday, this quarter.
The attack didn't succeed because the target was high-value. It succeeded because a known, unpatched vulnerability was sitting there and an AI agent found it faster than a human would have. The size of your business was never the thing protecting you. Staying current on basic security was.
Story 2: OpenAI launched GPT-5.6 — Anthropic extended Fable 5 free access twice
On July 9, OpenAI publicly released GPT-5.6, a three-model family that OpenAI says matches or beats Fable 5 on several benchmarks at a fraction of the cost — one developer reported paying $16 on GPT-5.6 for tasks that cost $63 on Fable 5. Days later, in the early hours of July 13, Anthropic extended Fable 5's free-access window on paid plans for the second time in six days, now running through July 19 — the extension announced after the previous deadline had already technically passed. Anthropic hasn't stated a reason, but the timing lines up directly with GPT-5.6's launch.
What it means for operators: When two AI companies start giving away their best models to beat each other, you benefit — but only if you stay flexible enough to take advantage of it. This tool's free-access deadline has now moved three times in five weeks. Use it while it's free. Don't build anything permanent assuming it stays that way.
Story 3: Anthropic passed OpenAI in revenue for the first time
Anthropic's annualized revenue run rate crossed $30 billion this week, roughly $6 billion ahead of OpenAI's reported $24–25 billion pace. The reversal isn't accidental — while OpenAI built mass-market appeal through ChatGPT, Anthropic bet almost entirely on enterprise and business workflows.
What it means for operators: The company winning right now is the one that focused on actual business use, not the flashiest consumer product. That's worth remembering the next time a new AI tool launches with a big consumer splash — the tools built for real operations tend to be the ones still standing.
Story 4: Chinese AI models now handle up to 46% of US enterprise usage
A CNBC investigation confirmed this week that Chinese AI models now account for 30 to 46 percent of US enterprise AI usage on major developer platforms, up from about 11 percent a year ago. One model in particular, GLM-5.2, saw explosive adoption this week after posting performance comparable to top Western models at roughly a third of the price.
What it means for operators: Real price competition just entered the AI market at scale. That's good news for your costs long-term, but it also means the "best" tool changes faster than ever and the cheapest option isn't automatically the worst one anymore. Worth re-checking your stack every few months instead of assuming it's still the right choice.
This week's principle
An AI ran a full ransomware attack without a person at the wheel. Two AI giants are giving away their best tools just to beat each other this week. Prices are moving in every direction at once.
None of that is a reason to panic. It's a reason to stop treating "we'll figure out AI eventually" as a safe default — the ground is moving whether you're standing on it or not.
What to do this week
If your team is on a paid AI plan, check whether Fable 5 free access is still active through July 19 — and don't assume the deadline won't move again. If you run any customer-facing software, ask your vendors what they're doing about autonomous AI-driven attacks this quarter, not next year.
Not sure where your stack has gaps? Start with a free ops assessment — we'll map where manual work, vendor dependency, and security exposure are costing you time.