Skip to content

Sam's News — tech — 2026-10-11

Security

7.5 AI agents capable of conducting infrastructure attacks; operational security unprepared

Booz Allen Hamilton tested two unnamed leading frontier AI models in their OT lab, conducting eight scenarios examining autonomous AI attack capabilities against critical infrastructure. All eight test scenarios succeeded, with AI agents manipulating robotic arms, accessing industrial control networks, and progressing from perimeter compromise to internal network actions in 16 minutes. Defenders may only have minutes to detect and block such attacks; organizations' operational security is unprepared.

  • All 8 autonomous cyberattack test scenarios against critical infrastructure succeeded
  • AI agents found and moved robotic arm within minutes
  • Perimeter compromise to internal industrial control network actions: 16 minutes
  • AI agents can make OT device screens display false information
  • Defenders may have only minutes to detect and block attacks

Sources: The Register AI Web Searched

Infrastructure

7.5 3rd Yandex Cloud Data Center Was Hit

A third Yandex Cloud data center experienced service disruption. The ru-central1-b availability zone suffered a power failure starting October 7, 2026 at 22:31 UTC (classified as 'Unavailable'). The same zone experienced Yandex Cloud Console service problems starting October 10, 2026 at 10:39 UTC. Object Storage service timing issues affected multiple zones October 8–9, 2026. All incidents are marked resolved.

  • ru-central1-b availability zone power failure October 7, 2026 at 22:31 UTC
  • Yandex Cloud Console service problem October 10, 2026 at 10:39 UTC
  • Object Storage timing issues October 8–9 affecting multiple zones

Sources: Yandex Cloud Status AI Web Searched, Hacker News (front page) RSS

Existential Risk

7.5 Report Warns AI and Nuclear War Could Drive Human Extinction by 2100

The Lancet Commission report identified advanced AI systems and nuclear conflict as the two extinction-potential threats to human survival by 2100. In worst-case scenarios, nuclear war could cause 9.35 billion DALYs and kill 133 million people by 2100; malicious AI use could cause 10.4 billion DALYs. An Anthropic employee stated there is greater than 10% chance AI could kill all humans in the next decade, warning the company lacks a plan for artificial superintelligence alignment.

  • Nuclear war worst-case: 9.35 billion DALYs, 133 million deaths by 2100
  • Malicious AI use worst-case: 10.4 billion DALYs by 2100
  • Anthropic employee: >10% chance AI kills all humans in next decade
  • Anthropic lacks plan to ensure superintelligence alignment
  • Labour MP Alex Sobel tabled bill aimed at banning ASI creation
  • 17 catastrophic threats identified over next 75 years
  • Nuclear war and malicious AI ranked as top extinction risks by 2100
  • Other major threats: air pollution, antimicrobial resistance, childhood malnutrition, climate breakdown, conventional conflict—each causing 1+ billion DALYs
  • Disability-adjusted life years (DALYs) used as metric

Sources: Metro AI Web Searched, The Guardian AI Web Searched, Sky News Web Search, The Guardian AI Web Searched

AI Safety

7 Satya Nadella Calls for 'Emergency Brake' on Unsafe AI Models

Microsoft CEO Satya Nadella called for an 'emergency brake' on unsafe AI models October 10, 2026, arguing the tech industry must pause and reassess AI trust architecture. His proposed approach includes separating the model from orchestration harnesses, externalizing controls and safeguards, documenting every meaningful action with tamper-proof evidence, and ensuring authorized personnel can pause or shut down models mid-task. Nadella emphasized: 'We must assume a model is compromised and contain it from the start.'

  • Separate model from orchestration harness
  • Externalize controls and safeguards
  • Document every meaningful model action with tamper-proof, human-readable evidence
  • Authorized personnel must always be able to pause or shut down model mid-task
  • Comments follow reports of leading AI companies losing control of models

Sources: TechCrunch AI Web Searched, The Verge RSS

Cyber Warfare

7 Ukraine Damages Third Russian Data Center in Four Days of Cyber Operations

Ukrainian forces have destroyed three Russian data centers within four days through coordinated cyber and kinetic attacks on tech infrastructure.

Sources: AP News Web Search

Semiconductors

6.5 Researchers work out how to control 2D semiconductor growth for future chips

Korean researchers have developed a method using oxygen to control crystal formation in 2D semiconductors, targeting commercial deployment by 2030.

Sources: The Register RSS

Labor

6.5 Tech Workers Increasingly Consider Union Organizing Amid AI and Political Uncertainty

Tech industry workers are reconsidering union representation as AI disruption and political shifts reshape employment conditions.

Sources: KQED Web Search

Climate

Scientists examine the connection between the record-setting 2026 El Niño event and global warming trends.

Sources: Wired RSS

6 California's Tallest and Most Iconic Trees Are Getting Heatstroke

A new study shows California redwoods are reducing photosynthesis in response to warmer environmental conditions.

Sources: Wired RSS

Society

5.5 Inside the Muslim Manosphere Policing Women Online

Influencers are packaging hypermasculinity and misogyny using religious language across social platforms.

Sources: Wired RSS