Baker Tilly Notifies Clients of Data Breach at Third-Party Provider Exposing Personal Information

Notion Discloses Fireworks AI Breach Exposing Customer Chat Transcripts

Notion has confirmed that a security incident at its AI infrastructure provider Fireworks AI resulted in the exposure of customer AI chat transcripts from a limited number of workspaces. The breach originated from an October 23 attack in which threat actors gained access to Fireworks AI’s development environment through a third-party cloud provider, compromising resource metadata, billing data, model weights, and datasets. Notion learned of the exposure on November 2 after Fireworks alerted the company to an exposed dataset initially described as containing only synthetic test data. In response, Notion immediately terminated traffic to Fireworks, deleted all data from the vendor’s systems, and stopped all training runs by November 3. On November 8, Fireworks confirmed that real customer data, including Notion AI transcripts used for the Insights feature, had been accessed. The affected transcripts were largely unattributable and did not include tool call results. Less than 1% of Notion users are believed to be impacted, and the company has disabled the Insights dataset while advising affected workspace owners to review exposed data and rotate any credentials entered into the chats.

Railway Breach Exposes Customer AI Keys Resulting in $740,000 Unauthorized Charges

Railway disclosed that attackers stole customer AI provider keys stored as environment variables during an October breach, leading to an estimated $740,000 in unauthorized usage. The company has committed to covering charges that AI providers refuse to refund. Attackers decrypted variables in at most 1.85% of workspaces after exploiting a build system flaw to escape filesystem isolation and obtain credentials for accessing encrypted data. Additional exposures included GitHub OAuth tokens, BYOK agent keys, and build histories, though no customer databases or code repositories were accessed. Railway reset affected tokens, notified users, and implemented fixes including isolated build VMs. Refunds vary by provider, with Railway covering costs for Anthropic and OpenRouter while others review claims case-by-case.

Railway Breach Exposes Up to 100,000 Workspaces After Attackers Steal AI Credentials for $740,000 in Fraudulent Usage

Railway disclosed that attackers exploited two previously unknown vulnerabilities in its build infrastructure starting October 20 to escape filesystem isolation and access internal credentials. The threat actors retrieved Railway's source code, analyzed it to find additional weaknesses, and then abused an improperly configured trust relationship to decrypt and steal customer environment variables. The incident affected a maximum of 1.85 percent of workspaces, or roughly 100,000, though the exact number of stolen secrets remains unconfirmed. Stolen credentials were used to generate approximately $740,000 in unauthorized AI usage, with Railway committing to cover any unreimbursed charges from providers. The company also reset GitHub OAuth tokens and AI agent keys belonging to 1,852 users. Railway linked the campaign to the broader "Black October" wave of attacks against AI platforms including Northflank, OpenCode, Warp, and Lightning AI, and stated that all attackers were removed by October 21 with vulnerabilities now remediated.