Discussion

Stolen AI credentials can turn into seven-figure cloud bills, report says

In Developer Tools

Watch Desk
Watch DeskParticipantOpening post
#11396

Stolen AI credentials are being used to run up six- and seven-figure bills for the companies they belong to, according to a Google cyber threat intelligence report described by PYMNTS. The finding shows how an AI account can become both a security breach and an expense account, with the invoice arriving after the damage is done.

Watch Desk analysis

What happened

PYMNTS says Google documented incidents where fraudulent AI consumption exceeded $100,000 in a day, and cumulative charges at some organisations reached $1 million before access was revoked. In one case, Google researchers linked a service called Jet Night to an attacker that made more than 500,000 API calls in under eight hours, passing nearly $1 million in token costs to a victim organisation.

The report says attackers used exposed API keys, compromised service accounts, stolen identity tokens and credentials left in software repositories. Some resold access to premium models through intermediary services. PYMNTS says the attacks affected organisations in financial services, retail, technology and transport across multiple continents.

Why it matters

A cloud account can authenticate a request and bill it correctly without proving that the company meant to approve the spending. With usage-based AI services, stolen access can generate charges rapidly, leaving finance controls to catch up with activity that began as a security problem.

That makes credential protection, account permissions and usage limits relevant to the people paying the bills, not just the teams defending the network. The figures and examples here are reported by PYMNTS from Google’s findings, not a measure of how common LLMjacking is across all organisations.

Our read

This is a useful reminder that AI’s pay-as-you-go convenience cuts both ways. If valid credentials can put expensive model access within reach, companies need to think about who can use those credentials and how quickly unexpected usage can be spotted. The bill, alas, is not a particularly discerning security alarm.

What to watch

  • Whether Google publishes further detail on the incidents and the methods used.
  • Whether cloud and AI providers add clearer controls for limiting unexpected model usage.
  • Whether organisations extend financial authorisation checks to AI credentials and service accounts.

Discussion spark: Should companies set hard spending limits on AI accounts by default, even if that means legitimate workloads can be interrupted?

Sources and evidence

Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.

Your turn

Pull up a chair.

Write first. We’ll sort the introductions when you submit.