A Flaw Triggered by a Wiz Alert and Northflank’s Immediate Response

On October 30, 2026, Northflank disclosed a security breach after an alert from Wiz detected suspicious activity in its infrastructure at 11:43 UTC. In a statement released the same day, the platform advised its clients to assume that all their environment variables had been decrypted, warning of a risk of theft of sensitive secrets. According to Northflank, clients reported unauthorized use of their AI API keys stored on its platform, confirming active exploitation of compromised credentials.

Northflank’s response was swift: by 13:00 UTC, the platform sought assistance from Wiz’s Cloud Incident Response Team (CIRT), deeming the incident’s scale beyond its internal capabilities. At 13:40 UTC, a public directive was issued to mandate the immediate rotation of all credentials and secrets, extended to all users by 13:58 UTC. This measure applied not only to AI API keys but also to database passwords (PostgreSQL, MongoDB, MySQL, Redis) and other critical identifiers stored in environment variables.

Northflank instructed its clients to assume that all their environment variables had been decrypted, a drastic measure highlighting the extent of the compromise.

A Context of Targeted Attacks Against AI Development Platforms

This breach follows a series of major incidents affecting AI development platforms in October 2026, described by Wiz as the « Black October for AI Developer Platforms ». According to Wiz’s analysis, nine platforms were impacted, including Fireworks AI, Zeabur, Railway, and Warp. Among these incidents, some involved Amazon Bedrock API keys used to generate 2.5 million unauthorized calls, resulting in estimated costs of around $100,000 for victims. Wiz researchers also observed dedicated infrastructures for reselling access to AI models, though without direct links to the breaches themselves.

Attack methods vary: credential theft, privilege escalation, or exploitation of vulnerabilities at third-party providers. In Northflank’s case, the platform did not disclose technical details of the compromise, but the attacks specifically targeted environment variables, often used to store sensitive identifiers such as database passwords, authentication tokens, or API keys. Some incidents, like those reported by Fireworks AI, OpenCode, and Warp, revealed the use of AI agents to accelerate attacks, a sign of evolving cybercriminal tactics.

What this means here: increased risks for Beninese and West African companies using cloud platforms and AI services

For businesses and government agencies in Benin and West Africa using cloud platforms like Northflank or externalized AI services, this vulnerability highlights several major risks. The compromise of environment variables, often underestimated, directly exposes critical credentials used to access databases, cloud services, or AI APIs. A similar exploitation could lead to unexpected costs, such as those seen with Amazon Bedrock, where unauthorized calls generated high bills without control.

Organizations should strengthen systematic rotation of API keys and secrets, even in the absence of a breach alert. Integrating anomaly detection solutions into usage and billing logs would allow for rapid identification of suspicious activities, as recommended by Northflank. Additionally, the widespread adoption of environment variable audits and restricting public access to exposed add-ons could limit attack surfaces. Finally, this incident underscores the importance of diversifying cloud and AI service providers to reduce dependency on a single platform, as Zeabur did by shutting down its services after its incident.

Public administrations, particularly those handling sensitive data, should also assess their exposure. Using dedicated API keys per service, rather than shared accounts, and enabling automatic revocation mechanisms in case of suspected compromise, could mitigate risks. Additionally, training IT teams in best practices for securing credentials and detecting LLMJacking techniques (such as API key theft for AI services) has become a priority.

A vulnerability that reveals structural gaps in credential security

Beyond the incident itself, this vulnerability highlights recurring weaknesses in credential management by cloud platforms. Environment variables, while convenient for storing configurations, are often insufficiently protected, as confirmed by Northflank, which recommends treating them as compromised. Attacks targeting these variables have allowed cybercriminals to access API keys for AI services like OpenAI, Anthropic, or Amazon Bedrock, used for queries to advanced models (Claude Fable 5.1, Grok, GPT-6).

Unauthorized calls began within the 30 minutes following the theft of the keys, with synchronized tests every minute, illustrating how quickly attackers exploit vulnerabilities. Northflank applied three network fixes the same day but did not specify how the systems were breached or how many clients were affected. This lack of transparency contrasts with the transparency required to reassure users, especially in a sector where trust is a cornerstone.

Northflank’s recommendations, disabling public access to add-ons, enabling audit logs for all clients (including free accounts), and generalizing credential rotation, could serve as a model. However, their implementation will depend on companies’ ability to integrate these measures into existing processes without creating operational bottlenecks.