Severity: HIGH | Actively Exploited in the Wild

Source: CISA Known Exploited Vulnerabilities (KEV) Catalog | CVE: CVE-2026-42208 | Affected Product: BerriAI LiteLLM

The Threat

The U.S. Cybersecurity and Infrastructure Security Agency (CISA) has officially added CVE-2026-42208 to its Known Exploited Vulnerabilities Catalog, confirming that this SQL injection vulnerability in BerriAI LiteLLM is being actively weaponized by malicious actors right now. LiteLLM is a widely used open-source proxy and SDK that allows organizations to connect applications to multiple large language model (LLM) APIs, including OpenAI, Anthropic, and others, through a single unified interface.

SQL injection remains one of the most destructive and well-understood attack classes in existence. In this case, an unauthenticated or low-privileged attacker can craft malicious input to manipulate the underlying database, potentially extracting sensitive data, bypassing authentication controls, or achieving full database compromise. CISA's addition of this CVE to the KEV Catalog is a hard signal: this is not theoretical. Exploitation is happening now.

Why East African Organizations Must Act Today

LiteLLM has seen rapid adoption globally as organizations rush to integrate AI capabilities into their products and internal platforms. In East Africa, this trend is accelerating. Kenyan fintechs, Ethiopian government digitization programs, Somali telecom providers, and regional banks are all building or piloting AI-powered services, including customer chatbots, fraud detection engines, document processing tools, and GovTech platforms. Many of these deployments use LiteLLM or similar AI gateway libraries as the backbone.

If your organization is running LiteLLM without the latest patch, your AI infrastructure, and the databases it connects to, may already be exposed. Given that Central Bank of Kenya (CBK) guidelines and the Kenya Data Protection Act 2019 impose strict requirements around customer data integrity and breach notification, a successful SQL injection attack could trigger regulatory penalties on top of the direct operational damage.

Impact Assessment for the Region

  • Financial Services (Kenya, Ethiopia, Somalia): Banks and fintechs using LiteLLM-powered AI for customer service or credit scoring risk exposure of customer PII, transaction records, and authentication credentials. This creates direct liability under CBK cybersecurity directives and CBE (National Bank of Ethiopia) data governance frameworks.
  • Government and GovTech Platforms: Public sector agencies in Kenya, Ethiopia, and Somalia piloting AI-assisted citizen services may be running vulnerable LiteLLM instances behind government portals. A breach here can compromise national ID data, civil records, or tax information.
  • Telecommunications: Regional telcos including those operating in Djibouti, Uganda, and Tanzania integrating AI into customer support platforms face risks of backend database extraction, including subscriber data and call records.
  • Healthcare: AI-assisted diagnostic or records management tools built on LiteLLM could expose protected patient health information, a growing compliance issue as East African nations develop health data regulations.

Immediate Actions - Do These Now

  • Audit your AI stack immediately. Identify every instance of LiteLLM deployed in your environment, including development, staging, and production. Check the version number against BerriAI's official security advisory and apply the latest patch without delay.
  • Isolate and restrict LiteLLM access. If you cannot patch immediately, place LiteLLM behind strict network access controls. Remove any public-facing exposure and whitelist only trusted internal IP ranges until a patch is applied.
  • Review database permissions. Enforce the principle of least privilege on any database accounts used by LiteLLM. A SQL injection attack's impact is directly proportional to the permissions of the compromised database user. Revoke unnecessary read/write/admin rights now.
  • Enable query logging and anomaly alerts. Turn on database query logging and set alerts for unusual query patterns, such as UNION-based queries, large SELECT statements, or access to system tables. This will help detect active exploitation attempts in real time.
  • Check for indicators of compromise (IoCs). Review your database and application logs for signs of prior exploitation, unexpected data exports, new database users created programmatically, or abnormal access times. Assume breach posture until you have confirmed a clean state.

DRONGO Recommendation

DRONGO's security team is actively monitoring CVE-2026-42208 across our regional threat intelligence feeds. If your organization is running any AI-integrated platform and lacks visibility into your full application stack, our Vulnerability Assessment and Penetration Testing (VAPT) service will identify exposed LiteLLM instances and other SQL injection vectors before attackers do. We operate across Kenya, Somalia, Ethiopia, and the broader Horn of Africa.

Is your organization protected? Request a free security assessment at DRONGO Technology Limited and let our regional experts confirm your AI infrastructure is not carrying an open door for attackers.