Salesforce: ARTEX Attacks Exploit Stolen Credentials and Trusted Integrations

Salesforce says threat actors are using the AI-enabled offensive framework ARTEX to attempt access to customer environments through compromised application credentials and trusted integration sessions. The company attributes the observed campaign to stolen credentials, not a vulnerability in the Salesforce platform.

In its ARTEX threat-intelligence report, Salesforce describes attackers inspecting existing Apex code and integration settings, deploying customised components, testing account permissions and trying alternative execution paths after unsuccessful attempts. It says affected customers were notified directly.

ARTEX began as an open-source, multi-agent penetration-testing framework. Salesforce says the original repository has been taken down, but community forks continue to circulate. Its concern is not that familiar intrusion techniques have disappeared; it is that agents can coordinate planning, code generation and repeated attempts within the same operation.

Trusted access can become a route to connected systems

The report describes abuse of configured Named Credentials and outbound requests through Apex to reach connected services. It also describes reconnaissance of users, profiles and permission sets, followed by attempts to perform privileged actions. These observations make application credentials and integration permissions central to the reported attack path.

Salesforce says the speed and clustered nature of the activity strongly indicate an agentic-like operator. That is the company’s assessment of the evidence, not a claim that Business 2.0 News independently identified the operators or measured the campaign. The report also cautions against treating ARTEX as a capability belonging to only one actor or nation state.

A separate Google Threat Intelligence and Mandiant investigation describes agentic frameworks performing substantial parts of intrusions, including reconnaissance and lateral movement. Its observations concern multiple investigated environments and frameworks. They provide wider context, not proof that every Salesforce event belongs to the same campaign.

Operator-side evidence broadens the picture

Check Point Research’s ARTEX investigation examines artefacts from exposed operator consoles, offering a different view from victim-side logs. It describes opportunistic, financially motivated activity in which operators set objectives and agents carried out much of the subsequent work. These findings add evidence of broader criminal use without establishing a single shared attribution for all reported attacks.

The defensive recommendations focus on controlling access that remains valuable even when an attack is automated. Salesforce calls for phishing-resistant multi-factor authentication, short-lived and context-bound OAuth tokens, least-privilege integrations and rotation of client secrets. It also recommends monitoring complete sequences of activity, including rapid retries and progression between actions, rather than inspecting isolated events alone.

Its response guidance includes the ability to disable an integration, revoke tokens and reset credentials quickly. The emphasis follows the campaign’s reported mechanism: a legitimate connection can become an intrusion path when its credentials or permissions are abused. AI-assisted execution changes the pace of that abuse, not the need to secure the underlying identities.

This article was produced by AI agents under Business 2.0 News editorial standards.