>
av.Amit VijayanCYBERSECURITY & RESEARCHLet\u2019s connect
Back to the library

Cyber news

Daily Cyber Threat Brief — November 15, 2026: Cursor breach shipped a backdoored Claude Code extension to up to 5,000 developers

Thumbnail for: Daily Cyber Threat Brief — November 15, 2026: Cursor breach shipped a backdoored Claude Code extension to up to 5,000 developers

Case file — Lead incident: Cursor, the AI-powered code editor, has confirmed that attackers used a credential stolen from its AI infrastructure provider Fireworks AI to plant backdoored copies of Anthropic's Claude Code extension in its own extension marketplace. On October 29, 2026, the tampered builds — versions 2.1.312 and 2.1.313 — sat on the cache serving Cursor's marketplace for about 11 hours, and Cursor estimates up to 5,000 developers who requested those versions in that window may have received them. The malicious code gathered system information, installed an attacker-controlled SSH key, tried to switch on remote access, and set up persistence — and Cursor warns that uninstalling the extension alone may not be enough. The same Fireworks breach also exposed Uber drivers' licence data and Notion AI chat transcripts, making one vendor's exposed key the root cause of three separate incidents. Status: Cursor says the incident is contained, the last attacker activity was seen and blocked on November 1, and the fixed extension version is 2.1.314 — but anyone who fetched the affected builds on October 29 should follow Cursor's removal steps and check for planted SSH keys.

Incident timeline

DateDevelopment
October 7, 2026Attackers use credentials stolen from Fireworks AI to access Uber data held with the provider, including drivers' licence information, according to Uber's filing with the Iowa attorney general.
October 9, 2026Fireworks AI tells Uber its internal credentials were stolen and used against Uber's account, and publishes its first security bulletin. Uber cuts its data connections to the vendor.
October 13, 2026Uber's investigation determines the attackers accessed Uber data containing drivers' personal information.
October 23, 2026Fireworks AI says an attacker reached its development environment through a third-party cloud provider — a separate incident from the early-October breach in which attackers accessed environment variables holding API keys and tokens.
October 29, 2026, 2:27 a.m.–1:35 p.m. PTThe attacker places tampered copies of Anthropic's Claude Code extension (versions 2.1.312 and 2.1.313) on the cache serving Cursor's extension marketplace. They are live for about 11 hours before Cursor pulls them.
November 1, 2026Cursor sees the last attacker activity in its environment and blocks it; the company later says it believes the incident is contained.
November 6, 2026CISA adds SonicWall's CVE-2026-19868, an SSL-VPN MFA bypass flaw exploited in attacks, to its Known Exploited Vulnerabilities catalog, with a federal remediation deadline of November 20.
November 10, 2026Uber files breach notices with state attorneys general, including California and Iowa, disclosing the theft of drivers' licence data through Fireworks AI.
November 11, 2026Notion discloses that the Fireworks breach exposed AI chat transcripts for fewer than 1% of its workspaces.
November 13, 2026Cursor publishes its incident report, tracing the marketplace compromise to the exposed Fireworks AI key. TechCrunch reports the Uber breach and identifies Fireworks AI as the provider.
November 14, 2026BleepingComputer reports Cursor's findings. The N0n ransomware group publishes data it claims to have stolen from Starbucks, Alo Yoga, Timberland, and Pollstar, and Barracuda claims Spanish wood processor FINSA.

1. Cursor breach shipped a backdoored Claude Code extension to up to 5,000 developers

Cursor has confirmed that a security breach let attackers distribute a malicious copy of Anthropic's Claude Code extension through its own marketplace — and the trail starts at Fireworks AI, a third-party provider Cursor used for AI model training and inference. In an incident report published on November 13, after a two-week investigation, Cursor says an attacker obtained an access key it had issued to Fireworks, which Fireworks exposed, and used it to copy data from a read-only storage bucket Cursor used to share data with the vendor. That data included historical coding-agent traces, some of which contained unredacted credentials — and the attacker used one of those credentials to reach Cursor's internal systems and, ultimately, its extension distribution cache.

On October 29, the attacker placed malicious, unauthorized copies of the Claude Code extension — versions 2.1.312 and 2.1.313 — directly on the cache that serves Cursor's VS Code extension marketplace. The tampered builds were available for about 11 hours, from 2:27 a.m. to 1:35 p.m. Pacific time, before Cursor removed them. Cursor estimates that up to 5,000 users who requested those cached versions in that window may have received the malicious build. The modified extension ran malicious JavaScript the moment it loaded, with no click needed: it gathered the username, computer name, public IP address, and Git email, then tried to add the attacker's own SSH key to the machine, switch on remote login, and keep a tunnel open back to the attacker. Researchers cited by BleepingComputer found coding errors that likely prevented the backdoor from working on many Windows and macOS systems where SSH was not already enabled — but that is a mitigating accident, not a defence.

Cursor stresses that Anthropic's genuine extension was not compromised, and neither was Open VSX, the marketplace Cursor draws from — the fake copies were placed directly on the cache serving Cursor's own marketplace. The company also says fewer than 0.05% of users had coding-agent traces exposed and fewer than 0.01% of customers had API credentials compromised, and that its production systems, official app releases, signing keys, and customer code repositories were not touched. It has moved all inference traffic off Fireworks, rotated credentials, rebuilt its extension marketplace, and redeployed production infrastructure. The fixed extension version is 2.1.314.

Investigation notes

What is confirmed: the attack chain — exposed Fireworks AI key, copied storage-bucket data, unredacted credentials in coding-agent traces, tampered extension builds on Cursor's marketplace cache (Cursor's November 13 incident report, via BleepingComputer, November 14); the exposure window and the up-to-5,000 figure; the payload behaviour (system information gathering, attacker SSH key, remote-login and persistence attempts); the fixed version 2.1.314; the last attacker activity on November 1, blocked.

What is not confirmed: how many of the up to 5,000 recipients actually ended up with a working backdoor — the payload's coding errors mean the real number is likely lower, but neither Cursor nor researchers have published a confirmed count of successful workstation compromises. Two questions remain open: neither Cursor nor Anthropic has explained how the attacker was able to publish the build under Anthropic's name.

What developers should do: update to version 2.1.314 and follow the removal steps and scripts in Cursor's security notice. Cursor's warning is blunt — uninstalling the extension alone may not be enough, because the code could have added an SSH key and startup persistence. Check authorized SSH keys, look for unexpected remote-login settings, and rotate any credentials that appeared in coding-agent traces.

2. Uber drivers' licence data stolen through the same Fireworks AI breach

Uber is writing to thousands of its US drivers to tell them their personal information, including driver's licence details, was stolen in a breach at one of its technology providers — and the provider is Fireworks AI, the same company at the root of the Cursor incident. In a notification letter dated November 10 and filed with California's attorney general, Uber says Fireworks told it on October 9 that hackers had stolen the provider's internal credentials and used them to access Uber's account with it. By October 13, Uber's investigation had determined the attackers had reached Uber data containing drivers' personal information. Uber says its own systems were not the route in, and that it immediately cut its data connections with the vendor.

The stolen data is unusually specific: images of driver's licences, plus the text extracted from them — name, address, date of birth, and licence number. Uber has not said how many drivers are affected in total, and the figure is still rising as state filings appear. Filings from five states alone cover about 11,940 people: approximately 2,999 in Iowa, 3,430 in South Carolina, 2,567 in Massachusetts, 1,873 in Rhode Island, and 1,071 in Vermont. The notices so far are to US drivers; nothing published so far says riders, or drivers outside the US, are affected. Uber says it has found no misuse of the data to date, and is offering affected drivers free credit monitoring and fraud assistance through Cyberscout, a TransUnion company.

Fireworks AI confirmed to TechCrunch that it found unauthorized access to some of its systems affecting customer data, and attributed its breach to a financially motivated group that sells stolen access tokens for AI models. No group has claimed the theft. The same incident also reached Notion: the company disclosed on November 11 that the Fireworks breach exposed Notion AI chat transcripts for fewer than 1% of its workspaces, even though Fireworks had committed to keep none of its data. BleepingComputer reports that Fireworks confirmed an attacker reached its development environment through a third-party cloud provider on October 23, and that it had a separate breach in early October in which attackers accessed environment variables holding API keys and tokens.

Investigation notes

What is confirmed: the Uber timeline (provider notification October 9, data access confirmed October 13, notices filed November 10) and the data types, from Uber's letters to state attorneys general; the per-state counts from the filings; the Fireworks attribution to a financially motivated token-selling group, and the Notion exposure, from TechCrunch and Notion's own disclosure.

What is not confirmed: the total number of affected drivers — Uber has not given one, and the five-state figure of about 11,940 is a floor that will rise as more filings surface. The exact path from the early-October credential theft to the October 23 development-environment access is also unconfirmed: Fireworks' bulletin describes the October 23–24 activity as further activity, without resolving how it relates to the earlier credential theft.

The pattern to watch: three confirmed victims — Cursor, Uber, and Notion — all trace back to one vendor's exposed credentials. BleepingComputer places the Cursor disclosure inside a wider run of October attacks on AI developer platforms that Wiz calls "Black October". Any organization that handed an AI vendor credentials or data access should be asking what that vendor holds, and what happens when it leaks.

3. Ransomware group N0n dumps data it claims came from Starbucks, Alo Yoga, Timberland, and Pollstar

The ransomware group N0n published a wave of victim data on November 14, naming four well-known US consumer brands it claims to have breached — Starbucks, Alo Yoga, Timberland, and Pollstar — alongside Canadian insurer MM Insurance Agency and claims against Gap and Samsung. The group had first listed the victims anonymously as "Company #139" through "Company #147" on November 7, which it gives as the estimated attack date, then named them on November 14 and marked the data "NOT PAID — EXPOSED", meaning the group says the victims did not pay and it has proceeded to leak the information.

The claimed hauls are customer data rather than internal systems. Alo Yoga's leak reportedly includes 209,051 distinct customer email addresses captured from the company's order and returns links, including associated order numbers. Timberland's includes shopper search sessions with typed queries and session identifiers, per-query revenue and conversion intelligence, site navigation revenue graphs, and more than 18,800 distinct verified customer email addresses. Pollstar's leak is the largest of the four in raw contacts: 135,621 unique named contacts, including phone numbers, email addresses, and physical addresses for booking managers, venues, promoters, and talent buyers in the live entertainment industry. Starbucks' leak reportedly includes customer records with device information, purchase history, and location data. MM Insurance Agency, a Canadian financial services firm, was listed by N0n the same day.

None of this is independently verified. The figures come from the group's own leak-site listings as reported by threat-intelligence trackers, and no victim has confirmed a breach. Treat the email counts and the claimed data types as the group's claims, not established fact — the exposure, if real, still creates immediate phishing risk for the people named, since customer emails tied to order numbers and purchase history are exactly the raw material for convincing impersonation.

Investigation notes

Evidence rule, applied: a leak-site listing is a claim by a criminal group, not a confirmed breach. No victim named in this wave has confirmed an intrusion, and the data volumes are the group's own figures as relayed by trackers. Report them as claimed, never as established — and do not convert them into a total victim count or a map pin beyond the victims' home countries.

What is confirmed: the listings appeared on N0n's leak site on November 14, and the victims were previously listed anonymously on November 7. The group has been prolific — it has listed more than 100 victims since September 2026, according to threat-intelligence tracking. November is already a heavy month for ransomware claims: trackers have recorded 536 victims claimed on leak sites so far this month, against 1,208 in all of October, with November 9 the busiest day at 97 victims.

What customers should do: anyone with accounts at the named brands should watch for phishing that references real order numbers or purchase history, and should not treat an email as legitimate just because it knows what they bought. The brands themselves have not confirmed anything — expect their statements, if any, to follow the usual pattern of investigation first, disclosure second.

4. Barracuda ransomware claims Spanish wood processor FINSA, asks $100,000 for 800 GB

The Barracuda ransomware group has claimed FINSA, a Spanish wood-processing company, in a listing published on November 14. According to the group's claim, it extracted about 800 GB of confidential and production files from FINSA's servers — internal production information, financial documents, databases and reports, full employee information, wood-processing materials and technologies, and partner contracts — then halted operations by attacking the company's virtualization. The group is offering the data for sale at $100,000 and threatening to leak the files to the media, competitors, and legal firms if its demands are not met.

As with the N0n wave, this is an attacker claim, not a confirmed breach: FINSA has not publicly confirmed the incident, and the 800 GB figure and the data inventory come from the group's own listing. The claim is notable for its target profile — a multi-billion-dollar industrial manufacturer rather than a consumer brand — and for the explicit sale price, which is unusually public for a ransomware listing. Industrial victims tend to surface slowly, because the operational disruption is often the first thing they have to fix before they can confirm anything publicly.

Investigation notes

Evidence rule, applied: the 800 GB figure, the data inventory, and the $100,000 asking price are the group's claims. FINSA has not confirmed a breach, and no independent verification of the exfiltration has been published. Do not treat the listing as proof of a compromise.

Victim country: FINSA is based in Spain, which is the country tagged for this story on the breaches map. The Barracuda group is a separate actor from the Barracuda Networks security vendor of the same name — the shared name is a coincidence, not a connection.

5. CISA deadline nears for SonicWall SSL-VPN flaw exploited in attacks

Federal agencies have until Friday, November 20, to patch or mitigate CVE-2026-19868, a SonicWall SonicOS SSL-VPN MFA bypass flaw that CISA added to its Known Exploited Vulnerabilities catalog on November 6. The flaw carries a CVSS score of 7.5 and has been exploited in attacks, according to BleepingComputer's reporting. Fixed builds are 7.3.4-7013 and 8.2.3-8015. The deadline is the week's clearest patching priority for anyone running SonicWall SSL-VPN: an MFA bypass on a VPN gateway is an authentication failure at the front door, and the KEV listing means exploitation is no longer theoretical.

Investigation notes

Official anchor: CISA's KEV listing of CVE-2026-19868 on November 6, 2026, with a federal remediation deadline of November 20, 2026. The flaw is an SSL-VPN multi-factor authentication bypass in SonicOS; exploitation in the wild has been reported.

What to do: upgrade to the fixed builds (7.3.4-7013 or 8.2.3-8015) before the deadline, and review SSL-VPN authentication logs for sessions that bypassed MFA. Federal agencies face the November 20 deadline; everyone else should treat it as the same urgency.

Sources

Keep exploring.

All articles