Today's definitive briefing on the historic Instructure data breach, the failure of US-EAST-1 cooling systems, and Anthropic's orbital compute alliance with SpaceX.
The hacking group ShinyHunters has claimed responsibility for a massive cyberattack on Instructure’s Canvas LMS, affecting nearly 9,000 schools worldwide. The breach involves 3.65 TB of data, including personal information of students and faculty.
Researchers have verified samples containing PII and internal credentials. ShinyHunters has threatened to leak the full archive by May 12, 2026, if ransom demands are not met.
A "thermal event" at an AWS data center in Northern Virginia triggered a major outage in the US-EAST-1 region. The disruption affected EC2 instances and EBS volumes in the use1-az4 availability zone.
Major platforms including Coinbase and FanDuel reported degraded performance. AWS noted that restoring cooling systems was "slower than anticipated" due to high-density hardware constraints.
Anthropic has signed a massive compute deal with SpaceX for "Colossus-scale" GPU clusters (300 MW+). The deal explores space-based compute capacity to bypass terrestrial power constraints.
The vision involves utilizing Starlink laser-mesh networking to distribute agentic services globally with minimal latency, powered by abundant solar energy in orbit.
An OpenAI-led consortium (NVIDIA, Microsoft, AMD) unveiled Multipath Reliable Connection (MRC). This protocol allows training on 100,000+ GPUs using two-tier Ethernet switches.
MRC eliminates "failure amplification" by handling packet loss in microsecond hardware loops, improving training efficiency for Mixture of Experts (MoE) models by 35%.
Telecommunications giant Telenor launched Telenor Sovereign Cloud, a standalone company focused on national data residency for critical sectors like energy and healthcare.
The entity ensures that Sovereign AI models are trained and hosted on localized hardware, fully compliant with GDPR and national security requirements.
The White House is drafting an executive order to **vet AI models** similarly to how the FDA reviews drugs, focusing on safety and reliability before public release.
Models exceeding **10^26 FLOPs** will undergo mandatory audits for CBRN risks and cyber-offensive capabilities conducted by the **U.S. AI Safety Institute**.
A supply-chain attack targeted **DAEMON Tools** installers, dropping backdoors on machines in the government and manufacturing sectors.
Attackers utilized valid but stolen certificates to bypass **SmartScreen**, enabling persistent exfiltration of industrial control data from restricted networks.
Get the daily technical briefing delivered to your inbox at 08:00 UTC.