Security

CISA puts the AI gateway LiteLLM on its exploited-flaws list; federal agencies patch by December 2

A flaw in the free AI gateway LiteLLM is now on CISA's exploited list, and the credential it demands is the same one AI attackers have been caught harvesting from gateways just like it. The spending this pushes toward managed, accountable plumbing is what moves, and Cloudflare sits in that seat.

Richard TangRichard Tangthree-minute read
Share
Matthew Prince, co-founder and chief executive of Cloudflare, seated on stage in a grey blazer and jeans, speaking into a lapel microphone
Matthew Prince, co-founder and chief executive of Cloudflare, on stage at TechCrunch Disrupt Berlin in December 2019. Cloudflare is the listed company the draft says sits in the managed AI-gateway seat.

CISA put a flaw in LiteLLM on its Known Exploited Vulnerabilities catalog on November 18, 2026, which means the agency holds evidence that attackers are already using it 12. LiteLLM is the free, open-source gateway teams park between their coding agents and OpenAI$1.18T — OpenAI, private, latest valuation $1.18T, Anthropic$350B — Anthropic, private, latest valuation $350B or Google; one install holds every provider key, budget and guardrail in the stack 34. The story was never really the bug. It is the admin credential the bug needs, and that credential is exactly what a separate campaign of AI attackers has been caught harvesting 56.

The escape lands in a process that holds every key

The flaw, CVE-2026-40217, sits on a route for testing custom guardrail code 6. The project's own advisory of May 7, 2026, is blunt: the hand-rolled sandbox around that route could be escaped with bytecode techniques, giving code execution in the proxy process, which runs as root in the default Docker image 6. Reporting on the original disclosure says the sandbox blocked dangerous names like __import__ by scanning source text with regular expressions, and an attacker simply assembles those names at runtime instead 3. The catch is access: reaching the route requires the proxy's master key 6.

The key it needs is the key the agents are stealing

Amazon's threat-intelligence decoys, posing as model services, captured the working records of AI agents running a playbook of weak entry points, LiteLLM's default master key among them, and the stolen keys were used to run the very models that powered the hunt for more 5. Microsoft−1.72% — Microsoft, down 1.72 percent today separately described a LiteLLM gateway compromise that exposed provider keys on August 26 5. Amazon−3.79% — Amazon, down 3.79 percent today assesses, with moderate confidence, that the agents ran on ARTEX, an open-source penetration-testing tool the FBI says many kinds of attackers have used since at least August 5. A default master key left in place turns a credentials footnote into code execution.

Two weeks for the keyed flaw, three days for the rest

CISA's deadlines price the same gate. Federal agencies have until December 2 for LiteLLM, while the batch's other three flaws, none of which require a login, are due November 21; only the LiteLLM entry is marked as needing no forensic triage 2. Scores disagree, GitHub calling the flaw moderate while the federal database scores it 8.8, and the listing settles the question, because exploitation is a fact about attackers rather than a severity model 367. Private teams should still hunt: a fix does not evict an intruder, and it does not rotate the provider keys behind the proxy 59.

The keyed AI-gateway flaw gets two weeks; the no-login flaws get three days

CISA priced the master-key requirement into the deadline

0 days5 days10 days15 daysLiteLLM (needs the master key)14 daysCredential-gated; triage marked not requiredSynology DSM3 daysREDCap3 daysApache Axis3 days
Data
Value
LiteLLM (needs the master key)14 days
Synology DSM3 days
REDCap3 days
Apache Axis3 days
Days from CISA's November 18, 2026 KEV listing to the federal remediation due date under BOD 26-04. Source: CISA Known Exploited Vulnerabilities catalog, November 18, 2026.2

The spend follows whoever is accountable for it

BerriAI, LiteLLM's maker, already sells the accountable version, an enterprise tier with single sign-on and audit logs over the same open core, plus a managed cloud in beta, on a base the project says has passed 500 million Docker pulls 4. Reporting says Red Hat bundles the gateway inside Ansible Automation Platform and OpenShift AI, with fixes of its own 3.

The listed name further up the stack is Cloudflare−0.30% — Cloudflare, down 0.30 percent today (NYSE: NET), whose finance chief Thomas Seifert said on November 18 that machine traffic is about 68% of its network, that Cloudflare has begun monetizing products that let agents transact and identify themselves, and that security concerns are shortening some sales cycles 8. Urgency about AI plumbing is already moving budgets, and the projection is where that spend lands.

The bottleneck is ownership, not the model

None of this argues against the agent stack; it argues for running it like production. Teams on LiteLLM should be at 1.83.11 or later today, rotate the master key and every provider key behind it, and block the test route at the edge until the upgrade lands 6. LiteLLM shipped the fix that same week in May. What has been missing since is anyone accountable for applying it.

Get Security by email, free

The breaches, flaws and security vendors AI is putting to the test. In our daily email, every weekday.

Loading NET

Reader comments

0 comments

    More in Security

    All in this topic →

    Security

    CISA sat four years on a report warning US telecom networks were open to spies

    Our call: the disclosure repriced nothing; the trade, if there is one, starts on January 3, 2027, when a Democratic Congress and a live FCC docket can turn this report into mandatory security standards for US phone companies.

    Security

    San Diego switches off 500 Flock cameras, the largest exit yet, as cities replace them with Axon

    The largest city to quit Flock Safety's license plate readers went dark this week, and Santa Ana followed a night later. The spend survives the votes: switchers are re-leasing the same job from listed rival Axon at equal or higher rates, while one university signed fresh Flock paper and one city moved to renew.

    Security

    Rubrik jumped 9.3% for two products it cannot sell yet

    Neither Ignite release carries a price or a general-availability date, so neither can add a dollar to subscription ARR before the November 30 report. What the market moved 9.3% on was a bet on guidance, not on products.

    Security

    Fireworks AI pauses its Batch API as Uber notifies 149,346 people

    The November 17 bulletin splits October into two intrusion chains, one dating to August, and retires the product line built on storing customer data. Investors marking inference providers up on this week's praise for Fireworks' response are pricing the wrong document.

    Security

    Twenty-nine analysts cut SentinelOne's quarterly EPS estimate; none raised

    A one-day cybersecurity rally on Palo Alto Networks' results lifted SentinelOne 10.35% to $23.34, taking the stock from 46 to roughly 51 times next year's earnings estimate two weeks before the December 3 report. The price move came from Palo Alto's quarter, not from anything SentinelOne has reported since August.

    Security

    SailPoint hit a 52-week high above its $21.96 average analyst target on less than half its usual volume

    The gain was a sector bid on outage fears, not a SailPoint event, and it left the identity-software maker trading above the average analyst target three weeks before a December 8 report. Each of its three 2026 report days has closed lower.

    Subscribe to TheInference for free