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.

BleepingComputerNovember 11, 2026Translated from English

Read at source

Notion has disclosed that a security breach at its AI infrastructure provider Fireworks AI exposed customer Notion AI chat transcripts belonging to a small number of workspaces.

The incident stems from an October 23 attack in which an attacker accessed Fireworks AI’s development environment via a third-party cloud provider. The compromise included resource metadata, billing information, model weights, and datasets for some customers. Fireworks AI had already suffered a separate breach in early October when hackers accessed environment variables containing API keys and tokens for a limited number of accounts.

Notion received an alert from Fireworks AI on November 2 regarding an exposed dataset that was initially reported to contain only synthetic test data. The company responded by cutting off all traffic to Fireworks AI, deleting all data from the vendor’s systems, and halting training runs by November 3. On November 8, Fireworks AI confirmed that real customer data had been accessed, including transcripts from Notion users.

Notion is a cloud-based workspace platform used for note-taking, document creation, internal wikis, project management, and task tracking. It incorporates an AI assistant capable of answering questions, summarizing content, and drafting text. Fireworks AI powers some of the open-weight models used by Notion AI and supports analysis of stored conversations for the Insights feature, which provides workspace administrators with visibility into how teams use AI.

The exposed transcripts originated specifically from the Insights dataset, which has now been disabled. Notion stated that the vast majority of impacted transcripts were unattributable, meaning they were not labeled with user or workspace information and did not contain tool call results. The company has published a FAQ page noting that Fireworks AI had contractually committed to zero data retention for Notion data.

Less than 1% of Notion users are affected. Impacted workspace owners can review exposed data through the Security incident review section in the desktop web interface and share download links with members. Notion recommends that affected users rotate any passwords, API keys, or other credentials that may have been entered into the exposed chats. Users who have not received a notification are likely not impacted.

Notion Fireworks AI Notion AI

ShareX Facebook LinkedIn Reddit

Meridian digestThe week's most consequential security stories, distilled into one briefing.See the digest

More in Breaches & Leaks

Breaches & Leaks

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.

The Hacker NewsNov 11, 2026Translated from English

Breaches & Leaks

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.

BleepingComputerNov 10, 2026Translated from English

Breaches & Leaks

Nitori Confirms Breach Exposing 37.62 Million Customer Records

Japanese furniture retailer Nitori disclosed a major data breach in which attackers accessed internal systems and potentially stole 37.62 million customer records. The intrusion was first detected on November 6, 2026, though evidence indicates the attacker had been present since mid-October. Exposed data includes names, email addresses, postal addresses, telephone numbers, dates of birth, gender, membership numbers, and order details for both individual and corporate customers. Account passwords and credit card information were not compromised in the incident. Nitori has notified Japan’s Personal Information Protection Commission and law enforcement, and is warning affected customers about possible phishing and scam risks. The company has blocked the unauthorized access, patched vulnerabilities, and enhanced monitoring to prevent recurrence.

BleepingComputerNov 10, 2026Translated from English