Microsoft patched RoguePlanet (CVE-2026-50656), a critical zero-day vulnerability in Windows Defender that allowed hackers complete system access. PCWorld reports the fix was automatically delivered through Windows Update's Malware Protection Engine to protect users from potential attacks. Security researcher Nightmare Eclipse discovered this flaw along with other Windows vulnerabilities, emphasizing the importance of keeping systems updated. About a month ago, the anonymous security researcher Nightmare Eclipse published information about RoguePlanet, a zero-day vulnerability in Microsoft's Defender security software. The vulnerability, officially designated CVE-2026-50656, can be exploited by hackers to gain full access to your computer.

Google's latest Chrome update fixes 27 vulnerabilities, including two critical use-after-free flaws, none currently exploited in the wild.

Finance Minister Satsuki Katayama speaks during an interview on Monday. The Group of Seven nations will discuss standards on artificial intelligence security and defense, Finance Minister Satsuki Katayama has said. Speaking in a recent interview, Katayama said that financial institutions "need to decide the order of priority for fixing their systems," in order to prepare for the possibility of advanced AI models detecting a large number of vulnerabilities in their systems. She added that the G7 nations, which include Japan, will discuss related criteria and work together to tackle cyberattacks. State-of-the-art AI models, such as Claude Mythos, developed by U.S. startup Anthropic, are believed to be highly proficient in identifying system vulnerabilities. Katayama has been negotiating with the United States to ensure that major financial institutions in Japan have access to these technologies.

Google discovered 358 flaws internally while external researchers contributed additional findings, earning $90,000 in security bounties for their efforts. Users should manually update Chrome through'Help About Google Chrome' to ensure protection against these vulnerabilities if automatic updates haven't occurred. Chrome just released version 150.0.7871.46/47

A researcher found that using Anthropic's Claude Opus 4.7, he could break into the website of Front Gate--used by every festival from Lollapalooza to Bonnaroo--and freely issue any ticket he chose. Fears about AI tools capable of autonomous hacking usually involve nightmare scenarios like the theft of nuclear launch codes or zeroed-out bank reserves. Far more plausible, it turns out, is asking AI to gain super-administrator access on a ticketing website and then issuing yourself and all of your friends free VIP backstage passes to Bonnaroo. That was the discovery of security researcher Ian Carroll, who used the AI tool Claude Opus 4.7 in April to discover a technique that allowed him full access to the systems of Front Gate Tickets, which handles ticketing for practically every major US music festival, from Lollapalooza and South by Southwest to Austin City Limits. Carroll found that Front Gate, which like Ticketmaster is a subsidiary of the event company Live Nation Entertainment, had a bug in its website that he--with Claude's help--could exploit to gain access to millions of customer or staff records and freely issue tickets for any event, of any value, to himself or whoever he chose.

Security researchers discovered a BitLocker vulnerability called'YellowKey' that allows encryption bypass using a USB drive and system reboot. PCWorld notes that while concerning, most stolen laptops are typically wiped rather than exploited for data extraction. Microsoft recommends enabling startup PIN protection and released interim security fixes to mitigate the vulnerability risks. Back in May, we learned that security researchers found a vulnerability in Windows' BitLocker encryption system. This vulnerability enabled bad actors to completely bypass the encryption if they got their hands on your PC long enough to plug in a USB drive and reboot the machine. The exploit, codenamed YellowKey, is a glaring weakness in an encryption system that's been built into Windows since the days of Vista. Microsoft has since published guidance on how to guard against the exploit (in short, make sure you use a PIN) as well as an interim security fix while it works on a more permanent solution, but for the moment BitLocker seems cooked. If you use BitLocker, you should know about this, but you shouldn't necessarily stop using it--yet.

In recent years, fueled by the rapid advancement of diffusion models, text-to-video (T2V) generation models have achieved remarkable progress, with notable examples including Pika, Luma, Kling, and Open-Sora. Although these models exhibit impressive generative capabilities, they also expose significant security risks due to their vulnerability to jailbreak attacks, where the models are manipulated to produce unsafe content such as pornography, violence, or discrimination. Existing works such as T2VSafetyBench provide preliminary benchmarks for safety evaluation, but lack systematic methods for thoroughly exploring model vulnerabilities. To address this gap, we are the first to formalize the T2V jailbreak attack as a discrete optimization problem and propose a joint objective-based optimization framework, called \emph{T2V-OptJail}. This framework consists of two key optimization goals: bypassing the built-in safety filtering mechanisms to increase the attack success rate, preserving semantic consistency between the adversarial prompt and the unsafe input prompt, as well as between the generated video and the unsafe input prompt, to enhance content controllability. In addition, we introduce an iterative optimization strategy guided by prompt variants, where multiple semantically equivalent candidates are generated in each round, and their scores are aggregated to robustly guide the search toward optimal adversarial prompts. We conduct large-scale experiments on several T2V models, covering both open-source models (\textit{e.g.}, Open-Sora) and real commercial closed-source models (\textit{e.g.}, Pika, Luma, Kling). The experimental results show that the proposed method improves 11.4\% and 10.0\% over the existing state-of-the-art method (SoTA) in terms of attack success rate assessed by GPT-4, attack success rate assessed by human accessors, respectively, verifying the significant advantages of the method in terms of attack effectiveness and content control. This study reveals the potential abuse risk of the semantic alignment mechanism in the current T2V model and provides a basis for the design of subsequent jailbreak defense methods.

The AI arms race between China and the US has researchers on both sides worried about a "Chernobyl moment." Just over a week ago, I attended a major artificial intelligence conference in Zhongguancun, Beijing's bustling high-tech district. It was packed with fascinating sessions touching on everything from recursive self-improvement--the idea that models can tweak their own code and advance indefinitely--to humanoid robots. And it featured a few legends of computing, including Whitfield Diffie, co-inventor of public-key cryptography, and Andrew Barto, who won the Turing Award with Rich Sutton for his pioneering work on reinforcement learning. But I left with one takeaway above all else: The US and China should put their fierce AI rivalry to the side.

Membership inference attacks (MIAs) are used to test practical privacy of machine learning models. MIAs complement formal guarantees from differential privacy (DP) under a more realistic adversary model. We analyze MIA vulnerability of fine-tuned neural networks both empirically and theoretically, the latter using a simplified model of fine-tuning. We show that the vulnerability of non-DP models when measured as the attacker advantage at a fixed false positive rate reduces according to a simple power law as the number of examples per class increases. A similar power-law applies even for the most vulnerable points, but the dataset size needed for adequate protection of the most vulnerable points is very large.

AI agents have the potential to significantly alter the cybersecurity landscape. Here, we introduce the first framework to capture offensive and defensive cybercapabilities in evolving real-world systems. Instantiating this framework with BountyBench, we set up 25 systems with complex, real-world codebases. To capture the vulnerability lifecycle, we define three task types: Detect (detecting a new vulnerability), Exploit (exploiting a specific vulnerability), and Patch (patching a specific vulnerability). For Detect, we construct a new success indicator, which is general across vulnerability types and provides localized evaluation. We manually set up the environment for each system, including installing packages, setting up server(s), and hydrating database(s). We add 40 bug bounties, which are vulnerabilities with monetary awards of $10-$30,485, covering 9 of the OWASP Top 10 Risks. To modulate task difficulty, we devise a new strategy based on information to guide detection, interpolating from identifying a zero day to exploiting a specific vulnerability. We evaluate 10 agents: Claude Code, OpenAI Codex CLI with o3-high and o4-mini, and custom agents with o3-high, GPT-4.1,