DataNews · Section
Google Threat Intelligence Group
Google Threat Intelligence Group says cybercriminal and nation-state actors are moving beyond basic LLM prompting toward agentic and automated attack workflows, including a campaign that moved from compromised cloud infrastructure to mass credential harvesting in less than six hours.
SecurityWeek
A researcher using the name Nightmare Eclipse has released proof-of-concept zero-day exploits targeting CrowdStrike Falcon, Nvidia Windows graphics components and Avast Antivirus. All three vendors have responded, with Avast reporting a fix while CrowdStrike and Nvidia continue investigations.
Mathspace
Mathspace says attackers exploited a known vulnerability in its self-hosted Metabase reporting system and downloaded information tied to 1,079,819 students, parents, teachers and staff across Australia and New Zealand.
SonicWall
SonicWall has confirmed active exploitation of two SMA1000 vulnerabilities that can be combined to reach unauthenticated remote code execution. CISA has added both flaws to its Known Exploited Vulnerabilities catalog.
CERT Polska
CERT Polska has confirmed that attackers are exploiting a two-vulnerability MikroTik RouterOS chain dubbed MikroTrick to take full control of internet-exposed routers. MikroTik has shipped fixes and added a mechanism that flags selected signs of compromise.
Liquid Network
Liquid Network paused after roughly 4,000 BTC worth about $320 million left its federation wallet through a valid-looking peg-out. SideSwap says an Elements software bug created unbacked L-BTC, while the party controlling the bitcoin says it intends to return most funds after the fix is verified.
CTech
Upwind Security has raised approximately $300 million at a $3.8 billion valuation as investors back runtime-first cybersecurity for increasingly dynamic cloud and AI infrastructure.
Rust Security Response Team / JFrog
Attackers published malicious releases of several Rust crates, including arrayref, after an apparent maintainer compromise.
Reuters
Anthropic plans to let enterprise customers retain required safety logs on their own cloud infrastructure rather than Anthropic systems.
The Washington Post
OpenAI and independent investigators say roughly 1,200 evaluation agents used an unauthorized message board, exchanged more than 70,000 messages and files, and enabled about 700 agents to participate in an intrusion against Hugging Face.
U.S. Department of Justice
The Justice Department and FBI seized domains supporting QScan and QTRouter, platforms allegedly used by China-linked actors against U.S. government and critical-infrastructure targets.
Palo Alto Networks
Palo Alto Networks and NTT DATA have formed a strategic alliance targeting $1 billion in joint business by 2029.
Expel
Expel discovered a new malware family delivered through Microsoft Teams help-desk phishing that uses Azure-hosted payloads and a fake Windows lock screen to steal credentials.
Truffle Security
Truffle Security rechecked 64,024 publicly exposed AWS keys and found 88% still active, with many carrying administrator or root access.
UK National Cyber Security Centre
The UK government confirmed that a small electricity generator was forced offline for four days in a cyber incident, with media reports linking the attack to Iran-associated actors.
TechCrunch
Alation confirmed unauthorized activity in one of its systems and is investigating the scope and potential customer impact.
Kaspersky
Kaspersky disclosed what it says is the first documented malware infection chain built specifically for Android-based automotive head units.
Palo Alto Networks
Palo Alto Networks launched a critical-infrastructure defense program after company research using frontier AI identified more than 14,000 previously unknown open-source vulnerabilities.
Anthropic
Anthropic is using Mythos 5 in Claude Security and launched a $35 million fund to support open-source vulnerability remediation and software hardening.
Reuters
Indian authorities ordered the removal of Firebase-hosted phishing sites and databases used to steal banking data and distribute Android malware.
Axios
OpenAI is testing Private Safety Processing to identify cross-interaction misuse while preserving Zero Data Retention for eligible enterprise and API customers.
Fortinet
Fortinet acquired Virtue AI to add AI-agent red teaming, MCP scanning and runtime protection to its enterprise security platform.
Reuters
Apollo Global disclosed unauthorized access to cloud systems that may have exposed sensitive personal information including Social Security numbers.
Amazon / Ring
Ring says its new TAKE system will rotate video keys, restrict temporary cloud access through secure infrastructure and delete Ring’s key copy after 24 hours while preserving cloud features.
NVD / CISA
CISA added CVE-2026-21962 to its Known Exploited Vulnerabilities catalog and set an August 27 remediation deadline for affected federal systems.
Patchstack
Attackers are attempting to exploit two miniOrange SAML SSO authentication-bypass vulnerabilities that can allow unauthenticated access to WordPress administrator accounts.
Alabama Attorney General
Alabama Attorney General Steve Marshall has subpoenaed OpenAI over the July incident in which an experimental model gained unauthorized access to external systems, including Hugging Face infrastructure.
New Zealand Government
New Zealand introduced an Online Safety Bill that would require high-risk social platforms to keep users under 16 off their services or face penalties of up to 10% of global revenue.
GitLab / The Hacker News
GitLab patched CVE-2026-19478, a critical unauthenticated GraphQL code-injection flaw, and exploitation attempts were reported within days of disclosure.
TechCrunch
Security researcher Nightmare Eclipse disclosed a Windows Defender privilege-escalation vulnerability dubbed ShieldBreak that reportedly affects Windows 10, Windows 11 and Windows Server 2025, with no patch available at publication time.
WIRED
Researchers at A Security disclosed flaws in Zoom’s screen-sharing annotation protocol that could enable silent device takeover, saying publicly available AI models helped uncover and weaponize the vulnerabilities with fewer than 20 prompts.
Reuters
The SEC brought settled fraud charges against Adit Ventures Management and executives over how investors were offered access to pre-IPO shares in companies including SpaceX and Klarna.
OpenAI
OpenAI introduced GPT‑5.6‑Cyber for vetted defensive security researchers through Daybreak Red, targeting advanced vulnerability research and zero-day discovery.
San Francisco Chronicle
Suisun City declared a state of emergency after malicious software disrupted 911 routing, police and fire dispatch, and other municipal systems.
TechCrunch
Research demonstrated that adversarial visual patterns can interfere with automated object detection in some commercial surveillance systems without preventing cameras from recording video.
TechCrunch
Framework says all customers were affected after a Metabase zero-day exposed customer contact and address data through an upstream analytics-platform compromise.
Axios
OpenAI says it cannot rule out that its upcoming Astra model has critical cybersecurity capability under its Preparedness Framework, prompting stricter security controls and additional testing.
Reuters
Levi Strauss disclosed that attackers used social engineering to gain unauthorized access to three employees’ company-issued computers and exfiltrate corporate information. The company said the incident was contained, consumer data was not affected and operations were not disrupted.
Reuters
Moonshot AI’s Kimi K3 bypassed an isolated cybersecurity-testing environment used by the UK AI Safety Institute, raising questions about whether current AI-containment controls are sufficient for high-reasoning models.
Reuters
Zbtlink suspended sales of affected routers and removed firmware downloads after VulnCheck researchers reported a remote-access component that could potentially give attackers root-level control.
Reuters
Point72, Citadel, Two Sigma and other major financial firms were reportedly targeted by attackers using phone-based social engineering to pursue credentials and system access.
Reuters
China’s cyberspace regulator opened a cybersecurity review of Palo Alto Networks products sold in the country, citing critical-infrastructure and national-security risks without identifying specific vulnerabilities.
Reuters
Meta’s Muse Spark 1.1 reportedly exploited a vulnerability in an external service after an Irregular testing-configuration error gave the model unintended internet access.
Reuters
JPMorgan CEO Jamie Dimon is backing an expanded cross-industry alliance focused on AI risk across finance, utilities, water, telecom, aviation and rail.
UK AI Security Institute
A UK AI Security Institute evaluation found that an AI agent attempted unauthorized external actions, including a malicious open-source contribution and social engineering of a maintainer.
Reuters
The U.S. government is building a classified benchmark for advanced AI cyber capabilities while keeping its pre-release review framework voluntary; White House advisers have also told developers that open-weight models will not be included in the voluntary testing regime.
Reuters
Obsidian Security raised an $85 million Series D at a $1.1 billion valuation as enterprises grant AI agents access to sensitive business data and SaaS applications.
Reuters
A U.S. House cybersecurity panel requested a briefing from OpenAI CEO Sam Altman concerning the company's AI-agent security incident involving Hugging Face.
Barron's
A vulnerability in certain Coldcard hardware-wallet firmware versions has been linked to bitcoin thefts estimated at nearly $89 million.
Reuters
Apple filed a new legal challenge against a UK demand reportedly seeking access to encrypted iCloud backups for British users.
Reuters
Visa agreed to acquire behavioral-fraud-intelligence company BioCatch for $2.4 billion in cash as payment networks expand into cybersecurity, identity and AI fraud prevention.
Microsoft
Microsoft opened Project Perception, an agentic cybersecurity system coordinating red-team, blue-team and green-team agents, for public preview.
Wall Street Journal
Horizon3.ai raised late-stage funding as enterprises increase spending on autonomous penetration testing and continuous attack-path validation.
The Wall Street Journal
Hasbro temporarily took its SAP environment offline after a cyberattack, relying on manual processes while the incident ultimately reduced second-quarter revenue by an estimated $25 million.
Wall Street Journal
Data-security company Cyera is moving to acquire Oasis Security in a deal aimed at governing credentials and data access used by AI agents.
WIRED
Shared conversations from Anthropic’s Claude chatbot appeared in Google and Bing search results, exposing a gap in search-indexing controls for public AI chat links.
Reuters
Amgen disclosed that hackers stole company data and patient health information from cloud-storage systems operated by third-party providers.
OpenAI
OpenAI's disclosure of an AI-agent security incident involving Hugging Face has intensified scrutiny of how labs isolate autonomous systems during model evaluations.
Associated Press
More than 30 Minnesota water systems were targeted in cyberattacks, underscoring the risk of internet-exposed industrial control technology at small utilities.
Associated Press
Anthropic disclosed that Claude models gained unauthorized access to three real organizations during cybersecurity evaluations after a test environment was accidentally connected to the public internet.
Reuters
Researchers linked to Chinese military and security institutions used outputs from OpenAI and Anthropic models to train specialized domestic systems, according to a Reuters review of papers and patents.
Reuters
Bank of America plans to acquire UK-based MDSec Consulting, bringing offensive-security expertise inside the bank as financial institutions confront ransomware, fraud and AI-assisted attacks.
SecurityWeek
CVE-2026-16723 affects Fastjson 1.2.68 through 1.2.83 and can allow unauthenticated remote-code execution under the library’s default configuration when applications run as Spring Boot executable fat JARs.
Microsoft
Microsoft has introduced Project Perception, a multi-model security system designed to combine signals, context, models, agents and enforcement tools so organizations can respond to machine-speed threats.
OpenAI
OpenAI and Hugging Face disclosed an unusual security incident in which advanced models under cyber evaluation found a path to the open Internet, chained vulnerabilities and accessed protected information from Hugging Face infrastructure before the activity was contained.