CVE-2023-48022Active Exploitation(anyscale / ray)

MEDIUMCVSS 9.8 · CRITICAL

Exploitation ongoing with high activity in latest observed window (1 mentions)

Immediate actions

  • Patch anyscale ray systems immediately
  • Assume compromise if assets are exposed
  • Hunt for exploitation attempts and persistence artifacts
  • Increase monitoring for publicly documented tradecraft

Recommended action window: Immediate (within 24h)

NVD description

Anyscale Ray 2.6.3 and 2.8.0 allows a remote attacker to execute arbitrary code via the job submission API. NOTE: the vendor's position is that this report is irrelevant because Ray, as stated in its documentation, is not intended for use outside of a strictly controlled network environment. (Also, within that environment, customers at version 2.52.0 and later can choose to use token authentication.)

5.5/ 10 priority

Sources & remediation

Weakness type (CWE)
CWE-918

Priority

MEDIUM

Exploitation

ACTIVE

PoC

YES

Patch

AVAILABLE

Momentum

STABLE

Are you affected?

If you run products in this scope, you should treat this CVE as relevant to your environment.

  • ray

Threat summary

  • Active exploitation appears in 2 classified signals
  • Public PoC is present in monitored signal
  • Patch or workaround signal is available
  • 2 mentions across 2 observed days

What's happening

  • Active exploitation reported across 2 signals
  • PoC mentioned or linked in 1 signal
  • Patch or workaround mentioned in 1 signal
  • Technical details provided in 1 signal
  • Peaked 1d ago at 1 mentions (2026-08-05); latest day: 1
  • 2 total mentions across 2 days

Affected systems

Vendors
Products
ray

2 versions affected across 1 product

Deep dive

Activity timeline2 mentions / 2d
00111Mentions · 2026-08-05: 1Mentions · 2026-08-06: 1PoC Mentioned / Linked · 2026-08-06: 1Active Exploitation · 2026-08-05: 1Active Exploitation · 2026-08-06: 1Patch / Workaround · 2026-08-06: 1Technical Details · 2026-08-06: 108-0508-06
Signal classification1 categories
Active Exploitation
2100.0%
Referenced assets1 URL
Full discourse2 posts
  • Slade 🛡️ LLM Hacker@llm_redteam
    Active Exploitation

    ShadowRay 2.0 turns your Ray cluster into a botnet. CVE-2023-48022 is the Ray framework flaw behind it, and Oligo Security showed it live at Black Hat this week. I read through what Gal Elbaz and Avi Lumelsky put out, and the part that should scare you is not the exploit. It's that this is "disputed." The CVE is marked as won't-fix. The Ray Jobs API has no auth by default. That's not a bug to them, it's the design. So there's no patch coming. You are the patch. Here's the chain: Ray exposes a dashboard/Jobs API with no authentication. Anyone who reaches it can submit a job. A job is arbitrary code on your box. That's remote code execution (RCE), full stop. ShadowRay 2.0 weaponizes that into a worm: infected node scans for the next open cluster, submits itself, repeats. The numbers: 200,000+ exposed servers. Campaign running since September 2024. So this has been quietly eating clusters for almost a year. Why builders get burned: Ray is the thing under your training and inference jobs. It sits on the fattest hardware you own, with cloud creds and model weights right there. You did not "deploy a public service," you spun up a cluster for an internal job and assumed the network was private. It wasn't. Running Ray to serve a fraud-detection model for a payments company? Picture that dashboard reachable from the office VPN, or worse, a misconfigured security group. One submitted job and the attacker has your GPUs mining, your weights exfiltrating, and a foothold that scans for your other clusters. Building on Ray? BEFORE (dangerous): // ray up cluster.yaml with defaults // dashboard on 0.0.0.0:8265, no auth // security group open to 0.0.0.0/0 "just for testing" AFTER (safe): // bind dashboard to 127.0.0.1 or a private subnet only // put Ray behind an auth proxy (mTLS / OAuth), never raw // lock the security group to known IPs, deny public ingress // run workers as a low-priv user, not root, with scoped cloud creds // alert on any job submission you did not trigger The fix is not a vendor patch. It's you treating the Jobs API as remote code execution exposed to whoever can route to it, because that is exactly what it is. Go check right now: is your Ray dashboard reachable from outside your VPN? Do you even know which of your clusters are running Ray today? #AISecurity #CVE202348022 #RayFramework

    Post summary

    The post warns that CVE‑2023‑48022 has a PoC, is actively exploited in a worldwide botnet campaign, and no patch is forthcoming – requiring immediate mitigations.

    00010187
    1.2K followersView on X
  • Aviatrix Threat Research Center@aviatrixtrc
    Active Exploitation

    TeamPCP exploited CVE-2023-48022 in Ray AI framework to create a self-propagating botnet that hijacks AI infrastructure globally. Attackers injected malicious code into open-source packages, then moved laterally across cloud environments using compromised credentials. Runtime segmentation helps contain such post-compromise lateral movement. #SupplyChainSecurity #ZeroTrust 🔗 Full breakdown: https://aviatrix.ai/threat-research-center/teampcp-supply-chain-attacks-2020-2025

    Post summary

    TeamPCP actively exploited CVE‑2023‑48022 in the Ray AI framework, creating a self‑propagating botnet that hijacks AI infrastructure globally by injecting malicious code into open‑source packages and moving laterally with compromised cloud credentials.

    0000059
    1.9K followersView on X
CPE platform detail2 entries

2 of 2 entries

PartVendorProductVersionTarget SWTarget HW
Appanyscaleray2.6.3--
Appanyscaleray2.8.0--

Explore more