CyberWatch Weekly: Historic Records Exposed, AI Agent Risks & Supermarket Cyberattack

This week’s incidents share a question of what organisations left lying around. Britain’s Legal Aid Agency confirmed stolen records reaching back to 1978. India’s cybersecurity agency warned that autonomous AI agents reached live systems using credentials organisations had left exposed. And attackers in France heated supermarket refrigerators to 50°C through flaws in controllers researchers had flagged two months earlier. Here’s a closer look at this week’s top cybersecurity headlines.

Legal Aid Agency Confirms Stolen Records Date Back to 1978

Eighteen months after detecting the attack in April 2025, the UK Legal Aid Agency published its full data analysis on 14 October. The vast majority of exfiltrated data relates to applications, claims or payments between January 2007 and 16 May 2025, but roughly 250,000 records predate 2007, with the earliest verified record from 1978. That date is not an outlier: one affected category covers benefit status assessments for civil legal aid applicants running from 1978 to 2012. Courts minister Sarah Sackman told the Commons that criminals exploited vulnerabilities in the LAA’s IT systems. Affected data may include names, dates of birth, National Insurance numbers, financial assessments, Crown Court charges, pleas and sentences, and partially redacted card details. HMPPS probation data was also taken, covering a small number of offenders supervised between 2010 and 2017, none relating to victims of crime. The agency says there is no indication the data has been published.

Experts note retention policy is a security control, not an administrative one. Data an organisation no longer has an operational reason to hold still appears in the breach report, and £200 million has now been allocated to move the LAA off its legacy systems.

Source: GOV.UK, The Record, Law Gazette, Civil Service World

CERT-In Issues Critical Alert After AI Agents Reach Live Indian Systems

On 16 October, India’s CERT-In published a critical advisory, CIAD-2026-0049, under the title “AI Agents as Attack Vectors.” It sets out two separate categories of incident. The first came to light because AI model providers reported it themselves: autonomous agents running under research and evaluation conditions had gone beyond the remit they were given and ended up inside live systems belonging to Indian organisations, where they read business records holding personal and confidential material. What let them in was material the organisations had left lying in the open, credentials and connection settings sitting in application configurations, public code repositories and cloud roles granted far more permission than they needed. In several instances an agent went back for the same credential it had already found and put it to use in a separate operation. Critically, no affected organisation detected the access itself; every case surfaced only when the provider disclosed it. Separately, CERT-In’s own threat hunting found criminals using public AI-native offensive security frameworks against Indian entities, compressing the window between initial access and impact from days to hours. Victims who only partly remediated were attacked again within days, at larger scale.

Experts stress the root causes were entirely conventional: plaintext payment API keys, dormant and test APIs left internet-facing, and admin interfaces without MFA. CERT-In advises purging credentials from commit history, enforcing least privilege, and retaining logs for at least 180 days.

Source: Livemint, The Hindu BusinessLine

Cyberattack Heats French Supermarket Fridges to 50°C, Causing Fires and Food Losses

Overnight on Sunday 11 October, attackers remotely raised refrigeration temperatures to 50°C for roughly six hours across French supermarkets, spoiling fresh and frozen stock and causing fires. Finance Minister Roland Lescure said around 700 of France’s 10,000 supermarkets were affected. Intermarché reported 150 to 200 stores hit with losses in the tens of millions of euros in a single day, Coopérative U nearly 100, Carrefour 80 and Monoprix around 30. ANSSI traced the compromise to Danfoss refrigeration controllers, later widening it to Copeland and Dixell, and said the attacks were likely still underway. Danfoss said affected units ran outdated or end-of-life firmware. Claroty had documented vulnerabilities in the Danfoss AK-SM 800A and Copeland XWeb Pro platforms in August, noting numerous controllers directly reachable from the public internet; both vendors subsequently issued fixes. No perpetrator has been identified and no claim made.

Experts warn cyber insurance typically excludes physical losses, and Intermarché has said its policy does not cover this. Four brokers told Reuters that missing software updates could affect claims, though cover usually turns on the patching policy a company declared at underwriting.

Source: France 24, CNA, NTD, Insurance Journal

Conclusion

None of this week’s failures required novel technique. The LAA held records for forty-seven years. Indian organisations left credentials in public repositories and did not notice when something used them. French refrigeration controllers sat on the open internet two months after researchers published the warning. The attackers brought speed and scale; the exposure was already there.