Sam's News — tech — 2026-08-11¶

Security¶

10 North Korean remote IT worker infiltrated US government agency, FBI confirms¶

The FBI confirmed that a North Korean operative successfully infiltrated a U.S. federal government agency as a remote IT worker, demonstrating state-sponsored cyber infiltration capabilities that exploit weaker vetting processes in government compared to stricter security clearance practices.

North Korean operative hired as remote IT worker at unidentified U.S. federal agency
FBI Deputy Assistant Director Todd Hemmen disclosed discovery July 28, 2026
Infiltration represents rare confirmed instance of sanctioned North Korean penetrating U.S. government agency
North Korea conducting multi-year campaign placing remote IT workers to generate regime income and steal IP

Sources: TechCrunch AI Web Searched, Federal News Network AI Web Searched

8 Gunra ransomware exploits Fortinet vulnerabilities targeting critical infrastructure¶

Gunra, a Conti-derived ransomware-as-a-service group, is exploiting known Fortinet and Schneider Electric vulnerabilities to breach critical infrastructure worldwide, with U.S. and South Korean agencies issuing joint warnings on August 10–11, 2026, as the group has listed 51 victims since April 2025 and demands ransoms exceeding $10 million.

Exploits CVE-2025-24472 (Fortinet FortiOS/FortiProxy) and CVE-2024-5559 (Schneider Electric)
51 victims listed since April 2025; primarily in South Korea, Brazil, Spain, Thailand, Hong Kong, Australia
Ransoms exceed $10 million with 5–7 day payment deadlines
Uses Impacket tools (psexec.py, smbclient.py, secretsdump.py) for lateral movement
Employs double extortion: data theft plus encryption; publishes unpaid victims' data
Launched formal RaaS operations in January 2026 with configurable ransomware builders
Linux variant has cryptographic weakness allowing encryption key recovery

Sources: The Hacker News AI Web Searched, The Record AI Web Searched, The Register RSS

A Zoom screen-sharing vulnerability allowed attackers to remotely execute code and fully take over participants' devices during calls across all supported operating systems: Windows, macOS, Linux, iOS, and Android. The flaw affected anyone on a call involving screen sharing—whether participants or the host—and could be exploited silently with no indication to or interaction from the victim. The vulnerability was discovered in early June using publicly available AI models; cybersecurity firm A Security found it required fewer than 20 AI prompts to uncover the vulnerabilities and create a working attack. A Security cofounder Omer Gull noted that previously such a discovery would have taken a team of five people approximately six months. Zoom has patched the vulnerability following disclosure by A Security.

Sources: WIRED AI Web Searched, 9to5Mac AI Web Searched, Engadget RSS

AI¶

10 ChatGPT and Gemini both reach 1 billion monthly active users¶

Google's Gemini reached 1 billion monthly active users, becoming Google's fastest-growing product ever, shortly after OpenAI's ChatGPT hit the same milestone in June.

Gemini is the 14th Google product to reach 1 billion users
ChatGPT hit 1B monthly users in June 2026, 1B weekly users in July
Gemini grew from 750M (Feb) to 950M (late July) to 1B monthly users
63% of Gemini users use voice; over 150M images generated daily
Gemini has 100M+ active iOS users

Sources: The Verge AI Web Searched, TechCrunch AI Web Searched

8 Anthropic adds invisible watermarks to Claude-generated text globally¶

Anthropic announced it will embed invisible, machine-readable watermarks in all Claude-generated text and attach signed provenance metadata to supported file types globally, complying with EU AI Act transparency requirements. The watermarks are designed to persist through copying and pasting, though heavy editing or format changes may strip them.

Applies globally to all Claude products (API, Claude, Claude Code, Claude Cowork, Claude Tag)
Text watermarks designed to persist when copied and pasted
Supported files (.svg, .png, .jpg) carry C2PA-standard signed provenance metadata
Policy applies to new models launched August 2, 2026 onward; retrofit underway for existing models

Sources: The Decoder AI Web Searched, Audacy AI Web Searched, InfoWorld RSS, Tech Times RSS, Euronews.com RSS, The Independent RSS

8 OpenAI expands Daybreak cybersecurity program with GPT-5.6-Cyber model¶

OpenAI expanded its Daybreak cybersecurity program with two access tiers and launched GPT-5.6-Cyber, a specialized model for security researchers that completed 95% of advanced vulnerability tasks compared to 1.5% for standard safeguards. The model has already identified multiple previously unknown vulnerabilities in major software systems.

GPT-5.6-Cyber completed 95% of advanced cybersecurity tasks vs. 1.5% for standard guardrails
Two-tier access: Daybreak Blue (defensive) and Daybreak Red (specialized research)
Discovered 2 zero-days in Chrome V8, 5+ in mobile OS, 3 critical in database, 400+ in OS kernel
Model reached High cybersecurity capability threshold under OpenAI's Preparedness Framework

Sources: Quartz AI Web Searched, TahawulTech.com RSS, gadgets360.com RSS, qz.com RSS, Digital Trends RSS, Infosecurity Magazine RSS

8 Anthropic's unreleased model makes progress on the Riemann hypothesis¶

Anthropic said an unreleased research version of Claude made measurable progress on the Riemann hypothesis, raising a key lower bound from 41.6% to 67.2%.

Riemann hypothesis posed in 1859, carries a $1 million unclaimed bounty
Claude tested 650 ideas across 60 sub-agents over about a day and a half
Used roughly 31 million (compute units) in total resources
Combined recent work by Baluyot, Goldston, Suriajaya, Turnage-Butterbaugh with a 2000 Bombieri paper
Produced a formally verifiable proof in Lean; confirmed by mathematicians Brian Conrey and Dan Goldston
Anthropic says the technique likely won't lead to a full proof of the hypothesis

Sources: TechCrunch AI Web Searched, Anthropic AI Web Searched

Climate¶

8 AI could increase fossil fuel industry productivity and carbon emissions by up to 5 percent¶

New research finds that artificial intelligence could increase fossil fuel industry productivity and global carbon emissions by up to nearly 5 percent, with AI-driven fossil fuel gains exceeding environmental impact of AI data centers themselves.

AI could boost fossil fuel productivity and increase global emissions by up to nearly 5%
AI-driven productivity gains will boost oil/gas production more than clean energy development
AI environmental impact on fossil fuels expected to outpace AI data center energy consumption

Sources: WIRED AI Web Searched, Financial Times AI Web Searched, Wired RSS

Legal¶

8 Meta's Section 230 defense insufficient to block $1.4 trillion state lawsuit trial¶

A unanimous 9th Circuit panel ruled Meta cannot use Section 230 to block a multi-state trial over alleged addictive design of Facebook and Instagram, calling it a liability defense rather than immunity from suit. Trial is set to begin August 19, 2026, with jury selection August 12.

3-0 ruling from the 9th Circuit Court of Appeals
States involved: California, Colorado, Kentucky, New Jersey
Meta says states seek over $1.4 trillion in damages
TikTok's similar appeal, which joined Meta's arguments, also rejected
Meta's emergency stay motion denied as moot

Sources: Ars Technica AI Web Searched

Funding¶

6 OpenAI completes $7 billion employee tender offer¶

OpenAI completed a $7 billion tender offer on August 10, 2026, allowing employees to sell shares and valuing the company at $852 billion, suggesting a potential IPO delay despite the company's confidential SEC filing in June.

$7 billion tender offer completed August 10, 2026
Company valued at $852 billion, matching March 2026 fundraising round
Employees can now sell shares without public offering
CEO Sam Altman noted missed internal goals in April 2026
Focus shifting to enterprise business strategy

Sources: TechCrunch AI Web Searched, CNBC AI Web Searched