Chinese Spies Use Commercial AI Tools to SUCCESSFULLY Break Into Companies

Like many AI tools, Claude has a code generating tool. You can use it for good. Or not so good.

Chinese spies are using the code generation tool to target companies. Tech companies. Finance. Chemical. And, yes, government agencies.

Anthropic created a paper documenting what they know. You can read it here.

Anthropic says this is the first documented case of AI agents SUCCESSFULLY breaking into target companies. These targets are considered high value to the Chinese and they want to collect intelligence from them.

Using the tool, they succeeded in some cases.

This just shows that hackers, like everyone else, are using AI to make their lives easier.

It also shows that they are getting better at getting around the AI guard rails, such as they are.

They used the AI code generator along with the open standard MCP (Model Context Protocol) to get the job done. MCP is the glue that connects the AI world to the “real” world.

The AI agents developed exploits and payloads.

Then a human spent between two and ten minutes confirming the results and then telling the AI GO! GO! GO!.

The agents found credentials, escalated privileges, moved across the network and then stole data. The human approved the theft of the data that the AI found. No sense wasting time stealing useless data, after all.

Anthropic detected an earlier, less sophisticated attack in August that hacked 17 companies and demanded ransoms for stolen data. In that attack, humans were much more driving the train. Not so much this time.

The only good news is that the AI hallucinated a bit so it made the attackers think they were more successful than they were. That does not mean they were not successful, just a bit less successful than the AI claimed.

The AI will get better meaning the attacks will be more successful with each passing month.

As a company you have two choices. Option one is to ignore the threat and continue business as usual. Option two is to harden your defenses. Option one is less expensive but also, likely, relatively soon, basically ineffective at stopping sophisticated hackers. Option two is harder and a bit more expensive and it involves significantly hardening your infrastructure.

Option two is NOT a “one and done”, it requires continuing effort, time and money.

Option two makes the attacks both less likely and less damaging.

But the choice is yours.

If you choose option two, please contact us.

Credit: The Register

by