“10 petabytes of data were allegedly stolen from a state-run Chinese supercomputer.”
Mercor and the LiteLLM supply‑chain blind spot
In April, Mercor — an AI startup described as working with organizations including Meta, Anthropic and OpenAI — reportedly lost four terabytes of data in a breach tied to a LiteLLM supply‑chain incident. Security experts quoted in the reporting framed the event as a cautionary example: connecting data to AI models via proxies such as LiteLLM can create "sensitive blind spots" that expose large data sets to downstream compromise.
The Mercor case sits alongside other incidents this month that trace their roots to third‑party code, tooling, and integrations rather than a single misconfigured server, underscoring the growing operational complexity when enterprises connect internal data flows to external AI components.
FBI surveillance system breach and early attribution
A breach of an FBI surveillance system was also reported in April, with potential exposure of criminal probes and the identities of targets of FBI surveillance. Early investigations, according to the reporting, suspected Chinese government‑affiliated hackers. The coverage linked this intrusion to a broader pattern of cross‑border operations, calling to mind previous large telecommunications compromises.
The item did not offer confirmed technical details of the intrusion but emphasized the sensitivity of the exposed material — criminal investigations and surveillance targets — and the early focus of investigators on state‑affiliated adversaries.
State‑run Chinese supercomputer: alleged exfiltration of 10 petabytes
Separately, a state‑run Chinese supercomputer was reported to have been hacked, with as much as 10 petabytes of data allegedly stolen. The compromised data were described as sensitive and could include classified defense documents, missile schematics, and other material of high consequence. The scale reported — 10 petabytes — ranks the incident among the largest single exfiltrations cited in the April roundup.
The allegation places national‑scale computing assets in the same threat landscape as commercial cloud and enterprise systems, heightening questions about how high‑capacity facilities are protected and monitored.
ShinyHunters claims: McGraw Hill and Medtronic incidents
The threat actor group ShinyHunters claimed responsibility for two high‑impact incidents referenced in April’s reporting. In one, a Salesforce database misconfiguration was reportedly exploited in a McGraw Hill breach; ShinyHunters claims to have stolen 45 million records. In another, ShinyHunters claimed responsibility for a breach of Medtronic's corporate IT systems. Medtronic, according to the coverage, maintained that other systems and operations were unaffected.
These cases illustrate two recurring vectors highlighted in the month’s reporting: classic configuration errors (Salesforce misconfiguration) and the public assertion of responsibility by known criminal groups.
Vercel, Booking.com, ADT, and the LAPD: customer data and municipal exposure
A cluster of incidents in April demonstrated the range of consumer and municipal exposure that can follow third‑party compromise. Vercel was breached through a third‑party AI tool that allowed an attacker to pivot into additional environments; the organization stated that a limited number of customers were impacted.
Booking.com reported customer booking information was breached, potentially exposing names, emails and phone numbers. ADT disclosed a breach that exposed names, phone numbers and addresses, and noted that it had found no evidence of payment information being exposed. The Los Angeles Police Department had sensitive records exposed following a breach of the L.A. City Attorney’s Office; seven terabytes of data were exposed, including witness names and unredacted criminal complaints.
Taken together, these incidents highlight that consumer‑facing records and municipal legal records alike can be amplified by vendor compromises and configuration failures.
ANTS: French government agency and the scale of citizen data exposure
France’s ANTS, the agency that stores citizens’ ID cards, passports and driving licenses, was reported breached in April. Initial reports suggested 19 million records were stolen; later reporting revised that estimate to a range between 12 million and 18 million records. The exposed records were described as sensitive personal data, reflecting the particular gravity when national identity systems are targeted.
The variance in reported totals illustrates the challenge of assessing scope quickly in large‑scale breaches of centralized identity repositories.
Across these ten incidents, three themes recur in the reporting: supply‑chain and third‑party tooling risks (LiteLLM and third‑party AI tools), configuration failures (Salesforce misconfiguration), and high‑scale exfiltration attributed to both criminal groups and suspected state‑affiliated actors. The practical question the month’s pattern leaves on the table — and one the reporting implies but does not answer — is which defensive changes organizations will prioritize when the threat picture spans startups, cloud services, national computing assets and municipal records.
Original reporting: https://www.securitymagazine.com/articles/102282-10-data-breaches-to-know-about-april-2026
