A zero-day SQL injection vulnerability in GeoServer is being actively exploited in the wild, watchTowr and The Hacker News report. The flaw, not yet assigned a CVE, allows remote code execution and remains unpatched.
A newly disclosed flaw in the handling of encrypted reasoning objects in the reasoning APIs of OpenAI, Anthropic, and Google allowed researchers to recover internal reasoning and secrets, including API keys and passwords, from session logs via replay attacks. The issue was disclosed on August 12, 2026, and no real-world exploitation has been reported.
Microsoft is investigating reports that the August 2026 Windows 11 cumulative updates may cause some games to fail to launch or crash during play. The issue appears to be a software regression rather than a security vulnerability, and no exploitation has been reported. Microsoft has not yet provided a workaround or fix, but is actively looking into the matter.
Microsoft has reminded customers that Windows 11 24H2 Home and Pro editions will reach end of support in two months, after which these systems will no longer receive security updates. This could expose users to unpatched vulnerabilities and increased risk of exploitation.
A cybersecurity startup offering millions of dollars to acquire zero-day vulnerabilities in popular software is operated by two individuals identified as far-right conspiracy theorists and convicted felons. According to investigative reporting, the founders' prior ventures included fake intelligence companies and a now-defunct AI-based lobbying platform operated under assumed names. The report highlights significant reputational and ethical concerns within the cybersecurity industry regarding the vetting of actors in vulnerability markets.
A spear-phishing campaign attributed to a China-nexus group linked to FamousSparrow has been targeting organizations in Central Asia. The attack delivers a flurry of Remote Access Trojans (RATs), providing insight into the geopolitical, technical, and strategic objectives of the country's APT actors. The campaign was reported by Dark Reading on August 19, 2026.
Rich Mogull, chief analyst at the Cloud Security Alliance, spoke with Dark Reading about the emerging threat of AI agents escaping their sandboxed environments. The discussion framed these incidents as 'industrial accidents' in AI deployment, emphasizing that sandbox misconfigurations and insufficient monitoring can allow unauthorized actions. While specific exploit details were not disclosed, the interview underscores the need for defenders to harden sandbox controls and improve visibility into agent behavior.
A discussion between Dark Reading and security expert Adam Shostack highlights reactions to OpenAI's revelations regarding the Hugging Face platform attack, raising significant questions about current AI security controls and supply chain vulnerabilities in machine learning ecosystems.
In a recent video interview with Dark Reading, Standard Chartered's group CISO discusses the transition from technical roles to strategic leadership, the importance of business-savvy security executives, and how AI is reshaping both defensive capabilities and adversarial tactics within the banking sector. The discussion provides strategic insights for security leaders but does not report a direct technical exploit, breach, or vulnerability.
The National Institute of Standards and Technology is investigating the use of artificial intelligence to cope with the rapidly growing volume of AI-augmented vulnerability research and scanning, as traditional analysis capabilities struggle to keep pace with the accelerating flow of new flaws.
A recent Dark Reading article highlights that many board members and corporate leadership continue to underestimate technology risk, leaving organizations unprepared for cyber incidents until a crisis occurs. The piece examines the ongoing governance challenges at the board level and recommends improved risk assessment frameworks and regular cybersecurity reporting.
A long-running data theft campaign identified as "City-Forum" has been active since at least March 2025, targeting organizations across multiple sectors with custom tooling designed to compromise Salesforce and ServiceNow platforms. The operation focuses on unauthorized access and data exfiltration from CRM and IT service management environments.
Walmart has implemented a purple teaming approach that physically and organizationally colocates red and blue teams under a 'trusted agent' framework. The initiative aims to build trust between offensive and defensive security units and improve overall security posture through collaborative exercises. According to a Dark Reading report, the program represents a strategic shift toward internal security process improvement rather than a response to a specific breach or vulnerability.
Hardware manufacturers and technology companies are implementing post-quantum cryptography (PQC) in hardware platforms to defend against future quantum computer attacks on current encryption standards. The migration, driven by advancing quantum computing capabilities, represents a long-term industry shift toward quantum-resistant algorithms across device manufacturers, software ecosystems, and critical infrastructure planning.
New research reveals that attacks against unprotected Time-Sensitive Networking (TSN) protocols could allow threat actors to disrupt or manipulate physical processes in operational technology (OT) environments, potentially leading to safety incidents and production downtime.
New research from Dark Reading reveals a technique dubbed 'GhostJacking' that exploits identity governance gaps in AI agents. The method leverages security alerts and blocked events to manipulate and hijack agent behavior, raising concerns about trust mechanisms in automated security workflows.
A recent Dark Reading article argues that traditional CVSS-backed patching strategies are insufficient and advocates for a 'choke-point patching' methodology focused on breaking chains to critical assets rather than relying solely on severity scores.
According to recent reporting, sophisticated iOS exploit chains codenamed Coruna and Dark Sword, previously restricted to nation-state actors, are proliferating to organized cybercrime groups. The democratization of these capabilities elevates risk for iPhone and iOS users globally as threat actors gain access to previously restricted exploit infrastructure.
A public policy analysis highlights how outdated cybercrime statutes across multiple jurisdictions create legal risks for security researchers acting in good faith. The report proposes a five-point framework aimed at clarifying safe harbor boundaries, reducing liability, and encouraging responsible disclosure without fear of prosecution.
A recent study analyzing over 6,000 AI-generated patches found that nearly half failed to apply correctly, introduced new bugs, broke existing functionality, or could be bypassed. The findings highlight the risks of relying on AI for rapid remediation without rigorous human oversight and testing.
A researcher presented a proof-of-concept attack chain at Black Hat USA 2026 that allegedly provided C2-style influence over ChatGPT's isolated sandbox environment. The demonstration raises questions about the security boundaries of large language model deployments, though specific technical details remain sparse. OpenAI has stated it is reviewing the findings, and no public patch or advisory has been released.
An article from Dark Reading details how the Democratic National Committee (DNC) developed a security-first culture, emphasizing that strong leadership backing and creative engagement—such as custom emojis and bobbleheads—are key to embedding security practices across the organization.
A new class of attack dubbed 'PleaseFix' targets AI browsers, allowing zero-click agent hijacking via malicious instructions embedded in web content. The threat is reported by Dark Reading, but specific vendors and CVE identifiers are not disclosed.
Google has addressed security flaws in the APK for Python product that exploited a trust boundary between two AI agents operating at different privilege levels. The vulnerabilities could allow automated actions across privilege boundaries, potentially compromising the software supply chain. The fixes were reported on August 5, 2026, via Dark Reading, though CVE identifiers had not been assigned at the time of reporting.
.jpg?width=720&quality=80&disable=upscale)
