Summary
- CISA has added CVE-2025-62593, a code injection vulnerability in Ray-Project Ray, to its Known Exploited Vulnerabilities catalogue.
- The vulnerability can be exploited remotely and is reachable via Firefox and Safari browsers, raising the risk profile for developer-facing infrastructure.
- Organisations using Ray as a development tool should assess internet exposure of any Ray instances immediately.
- CISA’s required action deadline is 20 August 2026, with mitigations to be applied in line with BOD 26-04 guidance.
- If no mitigations are available, CISA directs organisations to discontinue use of the product.
What Has Been Listed
CISA has formally added CVE-2025-62593 to its Known Exploited Vulnerabilities catalogue, confirming that the code injection flaw in Ray-Project Ray carries sufficient real-world risk to warrant mandatory remediation action for US federal agencies and strong guidance for all other organisations. The listing indicates active exploitation is either confirmed or the risk of exploitation is considered credible enough to act on.
What Ray Is and Why It Matters Here
Ray is an open-source distributed computing framework widely used to scale machine learning workloads, data pipelines, and other compute-intensive tasks. It is commonly deployed in developer and data science environments, which are increasingly part of production infrastructure. That positioning — sitting at the boundary between internal tooling and networked services — is precisely what makes this vulnerability consequential. Developer tooling is frequently less hardened than production systems, yet it often holds elevated access to code repositories, data stores, and compute resources.
Nature of the Vulnerability
The flaw is classified as a code injection vulnerability that permits remote code execution. CISA’s advisory notes that exploitation can occur through Firefox and Safari, suggesting a browser-mediated attack path. This is a meaningful detail for defenders: it implies that a user browsing to a malicious page, or interacting with a crafted resource, could inadvertently trigger the vulnerability against a Ray instance accessible from their environment. The exact technical mechanism beyond this is not detailed in the available source material.
Scope and Exposure Considerations
Organisations running Ray in any capacity — whether for AI and machine learning development, distributed data processing, or internal research workloads — should treat this as a priority review item. The advisory specifically highlights that developers using Ray as a development tool may be exposed. Security teams should map where Ray is deployed, whether those instances are accessible from the internet or from workstations running general-purpose browsers, and whether any mitigations or patches have been issued by the vendor. CISA places responsibility on each stakeholder to evaluate the internet exposure of their own assets.
Cloud Deployments Are Also in Scope
BOD 26-04, which governs CISA’s remediation guidance, includes specific provisions for cloud-hosted services. Organisations running Ray on cloud infrastructure are not exempt; the advisory directs them to follow applicable BOD 26-04 guidance for cloud services. This is worth noting for teams that may assume cloud-native deployments carry inherently different obligations under vulnerability management frameworks.
Remediation Deadline
The required action date under BOD 26-04 is 20 August 2026. For US federal civilian executive branch agencies, this deadline is binding. For others, it provides a useful benchmark. CISA’s remediation instruction is to apply vendor mitigations in accordance with vendor instructions, and to conduct forensic triage where required under the accompanying guidance. Where mitigations are unavailable, CISA’s direction is unambiguous: discontinue use of the product.
Why it matters
Ray is embedded in many organisations’ AI and data engineering pipelines, environments that are often under-governed from a security standpoint and may run with broad network access or privileged credentials. A browser-exploitable remote code execution path in such a framework creates a credible route for an attacker to move laterally into sensitive infrastructure. CISOs should not treat this as a narrow developer tooling issue — it is a supply chain and development environment risk that warrants the same scrutiny as any internet-facing system.
What to do now
- Identify all instances of Ray-Project Ray deployed across your environment, including cloud-hosted and developer workstations.
- Assess whether those instances are accessible from the internet or from machines running Firefox or Safari browsers.
- Apply vendor-issued mitigations in accordance with Ray-Project’s instructions as soon as they are available.
- Follow CISA’s BOD 26-04 patching guidance, including the specific provisions for cloud services if Ray is cloud-hosted.
- Where no mitigation is available, evaluate whether continued use of the product can be justified or whether it should be suspended.
- Conduct forensic triage of Ray instances as directed under CISA’s Forensics Triage Requirements if your organisation has BOD 26-04 obligations.
