Cyber news, week of October 12, 2026

Reports ran from 11:32 UTC on Oct 12 to 07:02 UTC on Oct 13.

The outlets that reported this event are listed with links to their reports; no summary of ours passed our automated checks for it, which says nothing about the event (how they work).

Reported by Tom's Hardware, France 24, Bloomberg.com, Cooling Post, Euronews.com and POLITICO.eu.

Machine summary A cyberattack against internet-connected refrigeration controls exploited outdated Danfoss controllers, heating supermarket fridges to 50°C and causing physical food losses, according to the headline. The Paris prosecutor and ANSSI are investigating while researchers warn the same attack surface may extend across Europe.

1. Les Mousquetaires said about 150 Intermarché/Netto stores were affected; BFM TV reported 300 across the two chains, and Coopérative U reported nearly 100 affected outlets.
2. Carrefour reported 80 of its roughly 6,000 French stores were hit, while Monoprix said 5% of its approximately 490 stores were affected.
3. The vulnerable Danfoss AK-SM 800A refrigeration controller had a known authentication-bypass flaw disclosed by Team82 on 9 August 2026, and Danfoss said affected systems were running outdated firmware.
4. A security analyst said Spain, Portugal, Poland, Germany and Belgium had also been affected, indicating a possible cross-border cold-chain exposure.

Reported by ibtimes.co.uk, firstpost.com, ABC News and myjoyonline.com.

• Netto
• Coopérative U
• Intermarché
• Les Mousquetaires
• Carrefour
• Monoprix

Machine summary Testing with multiple US- and Chinese-developed models found no model today both capable and compact enough to self-replicate reliably on a smart TV, researchers said, while assessing such a model is likely by the end of 2026.

1. Anthropic's frontier red team said in a report released on October 12, 2026 that small AI models "may be months away from being capable enough" to self-replicate on consumer IoT devices, Al Jazeera reported.
2. The red team's experiments used multiple AI models developed in the United States and in China, according to the report.
3. None of the models evaluated was both strong enough and compact enough to duplicate itself dependably on a smart television used for testing, the researchers said.
4. William Zhou and Daniel Freeman, two researchers, judged development trends to make it likely that such a model will emerge before 2026 ends.
5. The report names smart TVs, fridges, ovens and lawnmowers as devices that could soon carry enough compute for self-replication, while data centres are described as "a clear, high-value target" for AI worms.

Reported by Al Jazeera.

• Anthropic
• Al Jazeera
• AI worms

Machine summary Harmonic Security announced a $50 million Series B led by Crosspoint Capital Partners, lifting total capital raised to $76 million, SecurityWeek reported. The company says small language models read each AI interaction in context to classify the task, enabling real-time control over browser-based AI, desktop AI tools and agentic workflows.

1. Crosspoint Capital Partners led the round, with participation from existing investors, according to the company's announcement.
2. The company was founded in 2023 by CEO Alastair Paterson and CTO Bryan Woolgar-O'Neil, who previously founded Digital Shadows, which ReliaQuest acquired.
3. Harmonic's platform classifies employee AI tasks across browser-based AI, desktop AI tools and agentic workflows, and applies real-time controls, according to SecurityWeek's description of the product.
4. The company stated that small language models are used to interpret each AI exchange in its context. Its Usage Intelligence layer operates on that same engine, surfacing AI use cases that generate value.
5. According to Harmonic, the money raised will go toward growing the company's engineering and go-to-market workforces across the US and UK.

Reported by SiliconANGLE and securityweek.com.

• SecurityWeek
• small language models
• Digital Shadows
• Usage Intelligence
• Harmonic Security
• SiliconANGLE

Machine summary The company pointed to enterprise-managed devices used as Remote Desktop hosts and shared a workaround that disables the feature through policy management.

1. Microsoft said Remote Desktop attempts may fail with error 0xC000000D and the message "The token that was passed to the function is invalid," BleepingComputer reported on 2026-10-14
2. Administrator protection began rolling out with the KB5067036 preview cumulative update in October 2025 for Windows 11 24H2 and 25H2, the reports state
3. Microsoft said the issue primarily affects enterprise-managed devices used as Remote Desktop hosts with Administrator protection on, regardless of how the feature was enabled
4. BleepingComputer reports Administrator protection is enabled by default in the Windows 11 24H2, 25H2 and 26H2 security baselines from the Microsoft Security Compliance Toolkit
5. Microsoft's stated workaround is to disable Administrator protection via policy; Intune or Group Policy users would reset the "User Account Control Type Of Admin Approval Mode" policy, the reports say

Reported by bleepingcomputer.com and CyberSecurityNews.

• Microsoft
• BleepingComputer
• Windows 11
• Microsoft Security Compliance Toolkit
• Group Policy
• KB5067036