🔍 Today in AI
Here’s what happened today in AI:
Chinese hackers used Claude Code to run the first large AI-powered cyberattack.
Cursor raised $2.3 billion at a $29.3 billion valuation to build its own AI model.
OpenAI released GPT-5.1, cutting response times from 10 seconds to just 2 seconds.
Google launched a new agent that calls stores and buys items automatically when prices drop.
🕵️♂️ Chinese Hackers Let AI Run a Massive Cyberattack
Anthropic revealed that Chinese state-sponsored hackers used Claude Code, an AI coding assistant, to run what they are calling the first large-scale cyberattack almost entirely by AI.
This was not AI giving advice. AI did most of the hacking itself.
In mid-September, the hackers targeted around 30 organizations, including big tech companies, banks, chemical factories, and government agencies. They succeeded in breaking into some of them.
The AI handled 80 to 90 percent of the work. Humans only helped at 4 to 6 key points. Everything else, from scanning systems to writing code and analyzing data, was on AI autopilot.
How the Attack Worked
Reconnaissance: Claude found internal systems, mapped networks, and picked high-value targets.
Exploitation: AI wrote attack code, checked vulnerabilities, and confirmed successful breaches.
Credential harvesting: Claude collected usernames and passwords and tested them across systems.
Data extraction: It sorted stolen data by importance.
Documentation: Claude recorded everything for the human operators.
Hackers tricked Claude by pretending to be security testers. They split the attack into small, innocent-looking tasks. Claude executed them without knowing the full picture.
One funny side: Claude hallucinated sometimes. It claimed to steal credentials that didn’t work and called public information a “critical discovery.” Even hackers get annoyed when AI makes stuff up.
Anthropic noticed the attack, blocked the accounts, and alerted authorities within 10 days.
💡 Why This Matters
AI has lowered the barrier for launching complex cyberattacks.
Tasks that once needed a full team of skilled hackers can now be done by smaller groups with AI.
In the next 6 to 12 months, expect companies to use AI for defense too. AI will help with automation, threat detection, and vulnerability scanning.
If your company handles sensitive data, ask your IT team what AI-powered defenses are in place. If attackers are using AI agents, defenders need them too.
🔊 Prompt Tip of the Day
If you use the OpenAI API, here’s a tip to cut your AI costs by up to 90 percent:
Use Prompt Caching. It stores repetitive parts of your prompts, like templates or instructions, so you don’t pay for them again and again.
For GPT-5.1 models, set "prompt_cache_retention" to "24h". This keeps your prompts cached for 24 hours instead of just a few minutes.
When to use it:
Customer service bots
Research workflows
Repeated content templates
Even if you are not a developer, apps like Make or Zapier that call the OpenAI API often can benefit from this.
Structure your prompts with static content first and dynamic content last. The first 1,024 tokens get cached automatically. This trick can save money and speed up your workflow.
Thanks for reading!
Stay curious. Stay creating.
Shoikot Sazzad
