NEW Threat Research: When AI Gets Hijacked: Exploiting Hosted Models for Dark Roleplaying

[Krebs On Security] [Our Blog]
Illustration Cloud

Password spray enters the Okta-gon

Identity Providers (IDPs), like Okta have always been a juicy target for threat actors of all skill levels. Attackers often still find success with low sophistication techniques like password spraying. Permiso identified a large Okta password spraying campaign that took place in late August.

💡 All Permiso clients affected by this campaign have already been notified.

From August 27 - 31, a threat actor from the IP Address 185.241.208.110 with a user-agent of python-requests/2.28.1 attempted password spraying against about 50% of Okta clients that Permiso monitors. In this campaign, the threat actor successfully guessed passwords in multiple organizations, and in at least one case even passed an MFA check. Permiso did not observe any post-exploitation activity associated with this campaign.

While password spraying is not a new technique, the scale and success of this campaign makes it unique. Permiso recommends organizations that leverage Okta, to review all user.session.start events with a client.ipAddress of 185.241.208.110 . If the outcome.result is SUCCESS the threat actor successfully authenticated to the environment. Searching for the IP on its own will show all the attempts.

In order for the attacker to run this campaign, they needed to have usernames to attempt. Often times these are enumerated by LinkedIn scraping, but there are various ways this can be done. In this campaign the attacker did not bother to disguise their user-agent. While it can be somewhat noisy in some environments, reviewing for python-requests and python-urllib can be a decent signal.

Compromised credentials are involved in almost every breach. Monitoring your IDP is an important step to staying protected.

Illustration Cloud

Related Articles

When AI Gets Hijacked: Exploiting Hosted Models for Dark Roleplaying

Key Takeaways Attacks against GenAI Infrastructure like AWS Bedrock have increased substantially over the last six (6) months. Particularly with exposed access keys. Attackers are hijacking victim GenAI infrastructure to power their own LLM

Permiso Universal Identity Graph: Why Universal Identity Is Key to Solving Security Siloes

Permiso Security is excited to announce the release of the Universal Identity Graph, providing identity security risk visibility for all entity identities, including human and non-human, across all environments.

Introducing Azure Activity Log Axe: An Open-Source Tool to simplify and improve the analysis of Azure Activity logs

We are excited to formally announce the public availability of Azure Activity Log Axe, an open-source tool designed to simplify and improve the analysis of Azure Activity logs. The tool, initially unveiled at the fwd:cloudsec conference in

View more posts