Wiz warns of mass credential theft hitting AI developer platforms

Cybersecurity firm Wiz released findings detailing attacks on at least nine AI development platforms across October, identifying widespread theft and abuse of stored customer credentials.

The incidents affected platforms including Railway, Lightning AI, Fireworks AI, Warp and Northflank. Wiz researchers documented attacks where intruders specifically searched for stored environment variables holding AI service keys, database passwords and cloud credentials.

Once stolen, attackers used the credentials to make automated calls against major AI model providers at scale. In one case documented by Wiz, an attacker made roughly 800,000 calls using a single compromised access key, driving up more than US$100,000 in usage fees. Several affected platforms reported that the attacks appeared automated, with swarms of requests rapidly testing model quotas and permissions across multiple cloud regions.

One hosting platform, Zeabur, announced it will shut down entirely, citing the breach as a contributing factor.

Stolen keys sold on shadow markets

The activity forms part of a broader pattern security researchers call LLMJacking. Attackers compromise credentials for services such as Amazon Bedrock, Microsoft Azure OpenAI or Anthropic, then either consume the computing resources directly or resell access through informal broker networks to buyers looking for unmetered model usage.

Because development environments and automated pipelines often hold live administrative keys with high spending limits, a single exposed key can quickly result in substantial financial losses before an account administrator notices unusual activity.

What this means for Australian businesses

For Australian companies building internal tools or connecting business data to AI services, the campaign shows that securing credentials requires the same rigour as protecting online banking or customer databases.

Business owners should review where their staff and external contractors store AI access credentials:

• Check cloud and hosting environments. Audit platforms such as Railway, Northflank or custom cloud containers to confirm API keys and secret tokens are not stored in plain text or open configuration files.
• Set hard spending limits. Most AI providers allow account owners to set daily and monthly billing caps. Enabling hard caps ensures an unauthorised user cannot run up unexpected five-figure bills overnight.
• Rotate keys regularly. If your business used any of the affected developer services during October, revoke existing API keys for OpenAI, Anthropic, AWS or Google Cloud and issue fresh credentials.
• Restrict key permissions. Use least-privilege settings where available. Avoid issuing keys with unrestricted regional access or administrative permissions when a basic query key is all the project requires.
• Monitor usage dashboards weekly. Check account activity logs regularly for unexpected spikes in volume, unfamiliar model endpoints or access coming from unfamiliar regions.