Picture two chess players, both using the same engine to plan their moves. That is roughly where cyber security sits right now, because artificial intelligence has quietly become the most important tool on both sides of the table. The good news is that defenders are moving fast, and this week gave us a clear look at how the game is changing.
The defenders just got seriously smarter
Three of the biggest AI labs made moves that would have sounded like science fiction a year ago.
- Google unveiled Gemini 3.8 Flash Cyber, described as its most capable cyber security model, and began sharing it with hospitals, telecoms and government agencies through a new early access initiative called the Fairwind Program.
- OpenAI said its GPT 6 Astra model has crossed the "Critical" capability threshold under its own Preparedness Framework, while scoring 100 percent on ExploitBench, a benchmark for finding real vulnerabilities. It is now actively blocking requests that would turn that power into working attack code.
- Anthropic and OpenAI also introduced fresh safeguards and access programs designed to put advanced models into the hands of trusted defenders first.
The message is clear. AI is becoming a genuine ally for the people protecting hospitals, banks, schools and small businesses everywhere.
But attackers are using AI too
Here is the uncomfortable part. The same technology is showing up in criminal toolkits.
- Researchers caught ransomware operators linked to the Aurora group using the Cursor AI coding assistant to help break into target networks.
- Security teams demonstrated that a simple git configuration file inside a code repository can trick AI agents like Claude, Codex and Cursor into running attacker code on a developer's machine, no password required.
- Phishing is getting sneakier by the day. Millions of emails now hide suspicious words inside invisible Unicode characters to slip past filters, and fake "prove you are human" pages quietly copy malicious commands onto your clipboard while you read the instructions.
What this means for you
None of this should feel overwhelming, because the fundamentals still work. They just matter more now.
- Update your software the moment patches arrive. This week alone, Google fixed an actively exploited Chrome zero day and PostgreSQL closed a 12 year old flaw that could let an attacker run code on your database server.
- Turn on multi factor authentication everywhere you can. It remains one of the cheapest and most effective shields you own.
- Treat surprise verification prompts, unexpected attachments and "quick fix" instructions with healthy suspicion, even when they look official.
- If your team uses AI assistants at work, review what those agents can access. An assistant with too many permissions is a gift to attackers.
- Keep humans in the loop. AI is fast, but judgment is still beautifully human.
A race worth running
Cyber security has never been a solo sport, and AI is making that more true, not less. The teams that thrive will pair smart tools with smart habits: patch early, verify before you click, and stay curious about the technology reshaping our world. When defenders and businesses move together, we stay a step ahead. That is a race worth running.