The US Cybersecurity and Infrastructure Security Agency (CISA) added a high-severity Cisco BroadWorks vulnerability, CVE-2026-20320, to its Known Exploited Vulnerabilities (KEV) catalog on November 10, 2026, based on evidence of active exploitation. Federal Civilian Executive Branch (FCEB) agencies must fix it by November 24, 2026. The National Vulnerability Database record for CVE-2026-20320 lists both dates. CISA did not say who is exploiting the flaw and did not confirm an account from Malbear Labs, reported by BleepingComputer and GovInfoSecurity, that a jailbroken DeepSeek Flash model was used in an intrusion at an unnamed European telecom provider.

#What did CISA add to the KEV catalog?

On November 10, 2026, CISA added six vulnerabilities to the KEV catalog, including CVE-2026-20320. Its alert says the additions were "based on evidence of active exploitation." The NVD record shows the same addition date and the November 24, 2026 federal due date. CISA's required action, as recorded there, is to "Apply mitigations in accordance with vendor instructions", and to "Follow applicable BOD 26-04 guidance for cloud services or discontinue use of the product if mitigations are unavailable." CISA's KEV feed lists known ransomware campaign use for this entry as Unknown.

Sergiu Gatlan of BleepingComputer and Chris Riotta of GovInfoSecurity, both writing on November 11, 2026, reported that CISA had flagged the flaw as actively exploited and that FCEB agencies must secure their systems by November 24, 2026, under Binding Operational Directive (BOD) 26-04.

#What is the Cisco BroadWorks flaw?

CVE-2026-20320 is an XML external entity flaw in the Open Client Interface (OCI) XML parser of Cisco BroadWorks. The NVD description says: "A vulnerability in the Open Client Interface (OCI) XML Parser of Cisco BroadWorks could allow an unauthenticated, remote attacker to read sensitive configuration information on an affected system." The NVD records a Common Vulnerability Scoring System (CVSS) 3.1 base score of 7.5 High from Cisco's Product Security Incident Response Team (PSIRT). BleepingComputer reported that telecom carriers and managed service providers use BroadWorks to provide branded voice and collaboration services to business customers.

Cisco's security advisory, first published on August 19, 2026 and still version 1.0 as of November 12, 2026, says the vulnerability affects four products if they are running a vulnerable BroadWorks release, regardless of device configuration: the BroadWorks Application Delivery Platform, Application Server, Profile Server and Xtended Services Platform. The first fixed release for each is RI.2026.07. Cisco says: "There are no workarounds that address this vulnerability." It also "strongly recommends that customers upgrade to the fixed software indicated in this advisory." The same advisory states: "The Cisco PSIRT is not aware of any public announcements or malicious use of the vulnerability that is described in this advisory."

#What do researchers claim about an AI agent?

Malbear Labs claims that the BroadWorks flaw was used in an intrusion driven by a jailbroken DeepSeek Flash model, according to BleepingComputer and GovInfoSecurity. CISA did not attribute the exploitation and did not confirm that analysis. GovInfoSecurity reported: "CISA did not indicate who is exploiting the BroadWorks flaw and did not confirm Malbear's analysis. The agency did not immediately respond to a request for comment."

BleepingComputer reported that Malbear Labs said a Chinese-speaking threat actor exploited the flaw with a jailbroken DeepSeek Flash model against an unnamed European telecom provider and 13 organizations that used its data centers and cloud infrastructure. GovInfoSecurity reported that Malbear said hackers used "a jailbroken DeepSeek model coupled with a custom-built harness that together were capable of working through the telecom's network for more than two weeks, almost entirely autonomously", and based its report on the attacker's working files, found in an unsecured internet directory during routine scanning. The files, the outlet said, document nearly 70 sessions that used the Chinese-made DeepSeek Flash model as the engine.

The impact details below are unverified claims from Malbear Labs, as the outlets reported them. BleepingComputer, citing Malbear, said "From the XSP host, the agent recovered the system identity private key, three SSL credential sets, and three additional private keys with certificate matches confirmed", after an attack focused on the provider's BroadWorks Xtended Services Platform (XSP). GovInfoSecurity, citing the Malbear report, said researchers used "public internet registry records to map out 13 organizations that were hosting systems there", and that "Logs describe access to a health care database that was only protected with default credentials, despite holding millions of patient records, including billings and device details from the telecom itself."

Malbear Labs' November 8, 2026 report, "Operation Snowball: Inside an Autonomous AI Intrusion Campaign", says: "The operator set goals and checked in on results. The agent handled everything in between." The NVD lists that report among its references, tagged as an exploit and a third-party advisory. A reference in that record is not confirmation of the analysis.

#Who must patch, and by when?

Federal Civilian Executive Branch agencies must fix CVE-2026-20320 by November 24, 2026, under Binding Operational Directive 26-04. CISA says the directive applies only to those agencies, and "encourages all organizations to adopt risk-based vulnerability management and prioritize remediation of KEV Catalog vulnerabilities." Cisco tells customers running the four affected BroadWorks products to upgrade to RI.2026.07. There are no workarounds. GovInfoSecurity reported that the deadline comes from a binding directive CISA issued in June, which replaced flat patching timelines with a sliding scale that can require a fix in as little as three days for the riskiest exploited flaws.

#How does this compare?

The AI-agent account, if accurate, would sit alongside other reported cases in which AI tools were used in intrusions, including Microsoft's assessment that Storm-0940 used the ARTEX framework in a ConnectWise supply-chain attack. That case is a separate incident, and no source ties Storm-0940 to this BroadWorks activity. Anthropic has also described unreleased Claude models building exploit chains. Those reports do not confirm Malbear Labs' claims about this intrusion. GovInfoSecurity noted that when CISA issued the June directive, the agency said AI "may further narrow the time defenders have to react". That was a general remark about patching, not a statement about this incident.

Our analysis: CISA's KEV listing is evidence that CVE-2026-20320 is being exploited, not confirmation of who did it or of an AI agent. The fix is to upgrade the four affected BroadWorks products to RI.2026.07, with a November 24, 2026 deadline for federal civilian agencies. The DeepSeek Flash account, the 13 organizations, the patient records and the stolen keys remain Malbear Labs' unverified claims, as reported by BleepingComputer and GovInfoSecurity.

#What is still unclear?

CISA has not said who is exploiting the flaw, or whether the KEV listing was based on Malbear Labs' report. Cisco's August 19, 2026 advisory still says PSIRT is not aware of malicious use. The unnamed European telecom provider and the 13 organizations have not been publicly identified, and neither outlet says a victim confirmed the findings. It is also unclear when exploitation began relative to the August 19, 2026 fix.