Skip to content

AI/MLCritical 9.8

CVE-2026-77177

Remote code execution via prompt-injection in Open GenAI Stack (Jinja2 eval).

How scary is it?

updated 21 Nov 2026
  • Yes

    Works without logging in1

    No account and no user action needed (CVSS PR:N, UI:N).

  • Yes

    Reachable over the network1

    Can be attacked remotely (CVSS AV:N).

  • None known

    Public exploit code is out

    None on record in Exploit-DB, Metasploit, Nuclei templates or PoC-in-GitHub.

  • None known

    Attacks seen in the wild1

    None reported that we know of. CISA's SSVC assessment says exploitation: poc.

  • No

    On CISA's exploited list2

    Not in the catalog as of our last sync.

Scare-o-meter

Act this week

Serious and within reach of attackers. Schedule the fix this week.

  1. Low signal
  2. Routine patch
  3. Act this week
  4. Drop everything

Why: Reachable over the network with no login and no user action; CVSS 9.8.1 Read from the same rule that decides our alert emails.

Are you affected? only if all are true

  1. Product names from our AI analysisYou run one of the products named in our analysisOpen GenAI Stack (ogx-ai) 2026-06-11 / Meta AI backend (WhatsApp and other Meta services)

    We hold no version range from the vendor for this CVE. These product names were written by a language model from the CVE record: check them against the sources.

    See the products named

  2. An attacker can reach it over the network

    CVSS attack vector: Network (AV:N). Reachable over the internet or any routable network.

    Source 1

Read from the CVE record: affected versions and the CVSS vector. It narrows things down. It does not prove you are safe.

What to do

  1. Recommendation, AI-writtenIsolate affected instances, block/untrusted prompt input, disable unsafe template eval, apply vendor patch when…

    Isolate affected instances, block/untrusted prompt input, disable unsafe template eval, apply vendor patch when available.

    Written by a language model from the CVE record. It is not the vendor's wording: check it against the sources.

Steps marked AI were written by a language model, from the CVE record or from the linked news pages. They are not the vendor's wording: check them against the sources.

Major vulnerability. High-impact RCE in an open GenAI stack (ogx-ai) used by Meta (WhatsApp backend). Server compromise at scale and user-impacting services make this newsworthy.

CVSS
9.8Criticalscored by CISA
Exploited in the wild
Not knownpublic proof of concept per CISA SSVC; not in CISA KEV
EPSS
0.59%chance of exploitation in the next 30 daystop 51% of scored CVEsFIRST score of 21 Nov 2026
Fix
Not stated by the vendorno fixed release in the CVE record
Published
29 Sept 2026analysed here 30 Sept 2026

Summary

Impact

Remote code execution on backend servers via unsanitized Jinja2 expression evaluation from prompt-injected input, enabling full system compromise and data access.

Triage

Attack surface
Server-side
Exploitable
At runtime
Attack vector
network
Weakness
Can a WAF block it?
Partly. See why

The summary, impact and action text on this page are written by our AI analysis pipeline, or taken from the NVD description when a CVE was imported from NVD. They are not the vendor's wording: check them against the sources. A longer AI executive summary and technical analysis are generated for signed-in accounts on the Max plan.

Affected products

Products named in our analysis

  • Open GenAI Stack (ogx-ai) 2026-06-11
  • Meta AI backend (WhatsApp and other Meta services)

Fix and mitigation

Action required

Immediate action required

WAF and other mitigations

Can a WAF block it? Partly. Code injection payloads are often obfuscated, so they slip past generic WAF rules.

Also consider: remove eval sinks · RASP

Based on the weakness type only; no CVE-specific evidence yet. We check Cloudflare, AWS, Google Cloud Armor and Azure; for Akamai, Imperva, Fastly, F5, ask your vendor.

Remediation steps (AI-written)

Step-by-step remediation is generated for accounts on the Max plan, for CVEs that match a technology in their stack. Add this technology to your stack to see the fix checklist.

Exploitation evidence

CISA KEV catalog
Not listed as of our last sync.
Public exploit code
None on record in Exploit-DB, Metasploit, Nuclei templates or PoC-in-GitHub.
CISA SSVC
exploitation poc, automatable yes, technical impact totalAssessed by CISA 29 Sept 2026 (SSVC 2.0.3). Exploitation is none, poc (public proof of concept) or active.

Timeline

  1. 29 Sept 2026Published
  2. 29 Sept 2026CISA SSVC assessment: exploitation poc1 day later
  3. 29 Sept 2026CVE record last updatedsame day later
  4. 30 Sept 2026Analysed by vulneraibility.comsame day later

Common questions

Is CVE-2026-77177 being exploited?

Not confirmed. CVE-2026-77177 does not appear in CISA's Known Exploited Vulnerabilities catalog, which records only exploitation that has been observed and reported publicly. That is evidence of absence of a report, not evidence the flaw is unattacked. EPSS currently estimates a 0.59% probability of exploitation in the next 30 days.

How severe is CVE-2026-77177?

CVE-2026-77177 is rated Critical with a CVSS score of 9.8. Severity describes how bad exploitation would be, not how likely it is: pair it with exploitation evidence before deciding what to patch first.

Is there a patch for CVE-2026-77177?

Not known from our data. No patch reference has been recorded for CVE-2026-77177, which is not the same as no patch existing — a fix may have shipped without a tagged reference, or after this record was written. The vendor advisory is the only authority on whether a fix exists, and mitigations may be available regardless.

What does CVE-2026-77177 affect?

CVE-2026-77177 affects Open GenAI Stack (ogx-ai) 2026-06-11, Meta AI backend (WhatsApp and other Meta services). Confirm the exact affected versions against the vendor advisory before deciding you are exposed.

What should I do about CVE-2026-77177?

Isolate affected instances, block/untrusted prompt input, disable unsafe template eval, apply vendor patch when available.

Exploitation status reflects CISA's KEV catalog as we last synced it. Check the catalog directly.

References

NVDCVE.orgCVE DetailsSource

1 reference in the CVE record, grouped by the tags the record gives them.

Other links (1)

Where the data on this page comes from

Summary, impact and triage tags: written by our AI analysis pipeline from the sources above. Check them against the sources.

Similar CVEs

Get CVEs like this by email. Join the free digest: the CVEs we flag as major, all CVSS 8.0 or higher, in up to three emails a day (sent at 05:00, 11:00 and 15:00 UTC). Matching alerts to the products you run needs an account, and accounts are invite-only during the beta.