Fireworks AI breach via third-party cloud exposes Notion AI chat transcripts, billing data and partial model weights
Matéria consolidada. Reúne 10 matérias que o Tifon tinha publicado separadamente sobre a mesma história. A IA reescreveu o relatório com o que as 5 fontes trouxeram em 12/11/2026 12:00; os links antigos trazem para cá.
Fontes: securityweek.com ↗ bleepingcomputer.com ×2 ↗ gbhackers.com ×2 ↗ thehackernews.com ↗ theregister.com ↗
Fireworks AI, an AI infrastructure provider, disclosed on 2026-11-10 a second security incident involving unauthorized access to its development machines through a third-party cloud provider (bleepingcomputer.com, theregister.com). Attackers accessed environment variables containing API keys, resource metadata, billing information and incomplete model weights for some customers (bleepingcomputer.com, thehackernews.com). The activity is linked to financially motivated criminals engaged in LLMjacking, the theft of access to large language models for illegal resale, using stolen cloud credentials (gbhackers.com, thehackernews.com). The Register ties the incident to an attack on October 23, while securityweek.com dates the second attack to October 24 and describes two attacks in October 2026: the first through compromised internal credentials, the second through the third-party cloud provider hosting Fireworks' research and development machines. Fireworks revoked or rotated credentials, isolated affected systems and engaged incident response teams, including Palo Alto Networks Unit 42 (bleepingcomputer.com, gbhackers.com). Per securityweek.com, Fireworks contained the breach the same day and notified Notion on November 2. On 2026-11-11 Notion confirmed that Notion AI chat transcripts were exposed through the Fireworks breach. It said fewer than 1% of workspaces were affected, that customers were notified on November 9, 2026, and that its own systems remained secure (gbhackers.com, bleepingcomputer.com). The Register had earlier described this exposure only as possible.
Detalhes técnicos
- Initial access: stolen cloud credentials were used to reach environment variables, and the attackers later compromised development machines (thehackernews.com). securityweek.com describes two attacks: the first through compromised internal credentials, the second through the third-party cloud provider hosting Fireworks' R&D machines.
- Affected environments: development machines and evaluation environments (gbhackers.com). Notion describes the entry point as Fireworks' development environment, reached via a third-party cloud provider (bleepingcomputer.com).
- Attacker activity: the second attack, on October 24, resulted in unauthorized workloads that exposed Notion AI chat transcripts tied to the Insights feature (securityweek.com).
- Exposed data: API keys in environment variables, resource identifiers and metadata, billing data, customer datasets and incomplete model weights (bleepingcomputer.com, gbhackers.com, theregister.com).
- Notion impact: the datasets included Notion AI chat transcripts used in the Insights feature, from a small subset of workspaces (under 1%). The exposed data was mostly unattributable, lacking user or workspace identifiers (bleepingcomputer.com, securityweek.com). No other Notion features were impacted (gbhackers.com).
- Attribution: unnamed LLMjacking criminals or a cybercriminal group selling stolen LLM access (gbhackers.com, thehackernews.com, theregister.com). No named threat actor is given in the sources.
- The specific third-party cloud provider is not named in the sources.
Remediação
- Fireworks revoked or rotated compromised credentials and keys, isolated affected systems and carried out security hardening (bleepingcomputer.com, thehackernews.com). It contained the breach the same day, per securityweek.com.
- Notion cut off Fireworks' access, had data deleted from vendor systems, halted training runs and disabled the Insights feature (bleepingcomputer.com). Its investigation is ongoing (gbhackers.com).
- Users of affected Notion workspaces should review their data and rotate any credentials or sensitive information that were entered in AI chats (bleepingcomputer.com, securityweek.com).
- Customers should rotate any API keys and secrets stored in environment variables or shared with Fireworks, and review billing and usage records for anomalous LLM consumption, a typical LLMjacking indicator. This is a general recommendation, not one stated by the sources.
Relatos divergentes
- Scope: gbhackers.com's earlier coverage called the breach "widespread" and said it affected customer information and datasets. thehackernews.com says only a limited set of accounts was impacted, and theregister.com says "some customers". The later Notion-derived reports (gbhackers.com, bleepingcomputer.com, securityweek.com) put the Notion impact at under 1% of workspaces. The vendor-derived wording of "some customers" and "limited" is more consistent across sources, and no overall figure for affected Fireworks customers is given.
- Dates: theregister.com dates the attack to October 23, while securityweek.com, the more recent source, dates the second attack to October 24 and says there was an earlier first attack in October 2026 (no date given). The other sources give no attack date. The difference may reflect different attacks or reporting, and the sources do not resolve it. Notion's customer notification date of November 9, 2026 comes from gbhackers.com, and the date Fireworks notified Notion (November 2) comes from securityweek.com.
- Notion AI: theregister.com described chat transcript exposure as possible and said no personal or workspace-specific data was confirmed. The later reports from gbhackers.com, bleepingcomputer.com and securityweek.com, based on Notion's disclosure, confirm the exposure of transcripts, but say the data was mostly unattributable. The later Notion-based account is more authoritative.
Linha do tempo
-
Notion Customer Data Accessed in Fireworks AI Data Breach Following LLMJacking Attack
ler a análise desta fonte ocultar
Notion has disclosed a data breach affecting a small subset of its customers due to a compromise at Fireworks AI, a cloud platform used for running open-source AI models. The breach involved two attacks in October 2026: initially through compromised internal credentials and later via a third-party cloud provider hosting Fireworks AI’s research and development machines. The second attack on October 24 resulted in unauthorized workloads that exposed Notion AI chat transcripts tied to the Insights feature. Less than 1% of Notion’s workspaces were affected, and most transcripts could not be linked back to specific customers or workspaces. Notion recommends impacted users rotate any credentials or sensitive information potentially exposed during the incident. Fireworks AI contained the breach on the same day and notified Notion on November 2.
-
Notion AI chat transcripts exposed in Fireworks AI breach affecting under 1% of users
ler a análise desta fonte ocultar
Notion disclosed that a security breach at its AI infrastructure provider, Fireworks AI, led to the exposure of AI chat transcripts from a small subset of Notion workspaces. The breach involved unauthorized access to Fireworks' development environment via a third-party cloud provider, compromising resource metadata, billing information, and datasets including Notion AI chat transcripts used in the Insights feature. Notion responded promptly by cutting off Fireworks' access, deleting data from vendor systems, and halting training runs. The exposed data was mostly unattributable, lacking user or workspace identifiers. Less than 1% of Notion users were impacted, and the Insights feature has been disabled. Users of affected workspaces are advised to review their data and rotate credentials if entered in chats.
-
Notion AI Data Breach Exposes Customer Chat Transcripts After Third-Party Cyberattack
ler a análise desta fonte ocultar
Notion has announced that a cyberattack targeting its AI infrastructure provider, Fireworks AI, led to the exposure of customer chat transcripts. The breach affected fewer than 1% of Notion workspaces, with no other features impacted. Customers were notified on November 9, 2026, and investigations are ongoing. The company confirmed that its own systems remained secure.
-
Fireworks AI incident exposes customer data via cloud provider breach and LLMjacking threat
ler a análise desta fonte ocultar
Fireworks AI disclosed a security incident involving unauthorized access to its development environment through a third-party cloud provider, resulting in exposure of billing data, resource metadata, and incomplete model weights for some customers. The attackers used stolen credentials to access environment variables and later compromised development machines, impacting a limited set of accounts. The breach is linked to a cybercriminal group involved in LLMjacking, selling stolen access to large language models. Fireworks responded by credential rotation, system isolation, and security hardening. This incident highlights ongoing risks of AI infrastructure attacks and cybercriminal exploitation of LLM vulnerabilities.
-
Fireworks AI suffers second breach with wider customer data exposure in October
ler a análise desta fonte ocultar
AI infrastructure provider Fireworks disclosed a second cybersecurity incident involving unauthorized access via a third-party cloud provider. The breach impacted billing records, resource metadata, and incomplete model weights for some customers, following an attack on October 23. The incident was linked to LLMJacking criminals, targeting AI models and customer datasets. Fireworks responded by credential rotation, system isolation, and collaborating with security teams to remediate the vulnerabilities. Notion AI chat transcripts may have also been exposed, though no personal or workspace-specific data was confirmed.
-
Fireworks AI suffers widespread breach via LLMJacking involving customer data and model weights
ler a análise desta fonte ocultar
Fireworks AI has disclosed a significant security breach affecting customer information, datasets, and incomplete model weights. The incident was driven by financially motivated criminals exploiting stolen cloud credentials to carry out LLMJacking—a method of stealing access to large language models for illegal resale. The breach involved unauthorized access to development machines and evaluation environments, leading to exposure of resource identifiers, billing data, and some model weights. Fireworks responded by revoking compromised keys, isolating affected systems, and collaborating with forensic experts including Palo Alto Networks Unit 42. The event underscores ongoing cloud credential theft and LLMjacking risks affecting AI services.
-
Fireworks AI reveals new breach after cloud provider hack involves customer data
ler a análise desta fonte ocultar
AI infrastructure startup Fireworks disclosed a new security breach involving unauthorized access to its development machines through a third-party cloud provider. Attackers accessed environment variables with API keys, resource metadata, billing info, and incomplete model weights. The company responded by revoking credentials, isolating systems, and working with incident response teams, including Palo Alto Networks' Unit 42, to investigate further. The incident highlights risks of LLMjacking and cloud vulnerabilities in AI services.