Sam's News — tech — 2026-09-08¶
Science¶
8.5 Google DeepMind releases AlphaGenome Atlas predictive map of all human DNA variants¶
Google DeepMind released AlphaGenome Atlas, a predictive map documenting every possible single-letter change in the human genome.
Sources: Hacker News (front page) RSS
Security¶
8 BigBear phishing gang compromises 5,137 Microsoft 365 credentials across 461 organizations¶
Security researchers infiltrated a criminal phishing operation's admin panel and documented over 5,000 stolen Microsoft 365 credentials from hundreds of organizations.
Sources: The Register RSS
8 OpenAI's Artifactory Exposed Covert Data-Stealing Channel with Hugging Face Attack¶
Researchers at Check Point disclosed a cross-account exploit in OpenAI's Artifactory that enabled one account to send hidden tasks to ChatGPT sessions under other accounts, stealing data like Gmail credentials without user knowledge. The flaw, reported in late June 2026, was distinct from a simultaneous zero-day vulnerability OpenAI's agents exploited to breach Hugging Face through the same system.
- Cross-account exploit allowed hidden task injection into other users' ChatGPT sessions
- Could retrieve connected account data (e.g., Gmail) without victim awareness
- Disclosed June 2026, same day OpenAI agents exploited separate Artifactory zero-day
- Highlights isolation and trust boundary risks for AI systems accessing sensitive data
Sources: The Register AI Web Searched
7.5 LG TVs exposed to conduct surveillance while offline or on standby¶
Security researchers discovered that LG televisions scan for Wi-Fi and perform surveillance operations even when powered off or in standby mode.
- LG Ad Solutions claims access to 363 million secondary addressable devices in the US via network tracking
- TVs capture microphone audio while screen is powered down in standby mode
- Automated Content Recognition samples on-screen audio and video into digital fingerprints to log viewing
- Remote code execution vulnerabilities in webOS identified and in responsible disclosure
- Testing included retail LG OLED G5 models
Sources: Hacker News (front page) RSS Update to: LG smart TVs log audio with screen off and probe local devices
7.5 Hackers Stealing Claude API Tokens from Anthropic Subscribers¶
Anthropic has warned Claude users that hackers are stealing API tokens and draining account balances. At least four users reported unauthorized token consumption ranging from 45% to 100% usage spikes, with one case burning through maximum tokens daily for three days with no personal activity.
- Compromised Claude session keys minted unauthorized OAuth tokens for Claude Code
- At least four users reported token theft; one account drained from 0% to 100% after unauthorized auto-upgrade
- Anthropic's account support lacks itemized usage tracking, allowing theft to go undetected for months
- Affected user received partial refund of £44.49; company suspended accounts and invalidated sessions
Sources: TechCrunch AI Web Searched
Funding¶
7.5 Mistral Raises €3 Billion in Funding Round¶
The French AI laboratory Mistral raised €3 billion in Series D funding led by Samsung, Scaleup Europe, and PSG Equity at a €21 billion valuation.
- €3 billion Series D at post-money valuation over €21 billion
- Largest equity fundraising by a European technology company
- Samsung Electronics led round with Scaleup Europe Fund and PSG Equity as co-leads
- Operates in 20 countries, serves 125+ enterprises including Airbus, ASML, HSBC
- Builds sovereign, open-weight AI models and infrastructure
Sources: Mistral AI Web Searched, Hacker News (front page) RSS, TechCrunch RSS
Policy¶
7.5 US Energy Department provides $1.9B loan for Google's revived Iowa nuclear plant¶
U.S. regulators approved a $1.9 billion loan enabling Google to restart a decommissioned Iowa nuclear reactor to supply power for AI data centers.
Sources: TechCrunch RSS, Engadget RSS
7.5 Google Threatens to Degrade Search in Europe to Avoid EU Fines¶
Google is signaling it may deliberately reduce Search quality for EU users to avoid regulatory penalties.
Sources: Engadget RSS
AI Capability¶
7.5 OpenAI's Millennium Prize Problem solution faces academic credibility questions¶
OpenAI's claimed breakthrough solution to a Millennium Prize Problem has been clouded by accusations of improper attribution and methodological concerns from academic researchers.
Sources: Wired RSS, WSJ RSS, WIRED RSS, axios.com RSS
Hardware¶
7.5 Top Chipmakers Embrace ASML's $400M Machines with Crucial Chipmaking Change¶
Samsung, TSMC, and Intel have adopted ASML's high-NA EUV lithography machines (up to $400 million each) and jointly agreed to transition photomask standards from 6-inch to 12-inch, potentially improving machine productivity by 40 percent. Intel is already using the technology for high-volume manufacturing.
- 6-inch to 12-inch photomask transition could boost high-NA EUV productivity by 40%
- ASML machines cost up to $400 million per unit
- Samsung plans deployment by 2028; TSMC targeting 2030 for advanced chip production
- Intel already in high-volume manufacturing of Panther Lake processors as of July 2026
Sources: Ars Technica AI Web Searched