Researchers Use Anthropic AI to Exploit macOS Security Flaws
Security researchers from Calif utilized Anthropic’s Mythos AI to bypass Apple’s Memory Integrity Enforcement on macOS in just five days. By chaining two vulnerabilities, they achieved root access, highlighting the escalating threat of AI-assisted cyberattacks. While human expertise remained essential, the incident demonstrates how AI can accelerate exploit development against robust security architectures. Apple is currently reviewing the findings and developing a fix. This event has intensified debates on AI regulation and national security, prompting the White House to reconsider its approach to advanced AI models.
Editorial summary awaiting refresh
Cross-source coverage
Wire timeline
Researchers Claim Anthropic’s Mythos AI Assisted in Cracking macOS Security
Cybersecurity researchers from Calif, a Palo Alto-based firm, claim to have bypassed Apple macOS security using Anthropic’s new AI model, Mythos. The team developed a privilege escalation exploit that links two separate bugs to corrupt memory and gain unauthorized access to restricted device parts. While the discovery process took five days, the researchers emphasized that the AI tool alone was insufficient; human expertise was crucial to combine the vulnerabilities with other techniques. This incident follows earlier reports of Anthropic’s Claude model identifying numerous bugs in Firefox. Apple stated it is seriously reviewing the findings to ensure user security. Anthropic, which limited Mythos access to select partners, has warned about the severe consequences if such powerful vulnerability-finding capabilities fall into malicious hands. Experts remain divided on the implications; Google researcher Michał Zalewski noted the hype may be overblown but acknowledged the tool's utility for auditing, while others argue that restricting such technology does not solve underlying security problems. The report highlights the growing intersection of artificial intelligence and cybersecurity vulnerabilities in major operating systems.
PCMag.com - Technology Product Reviews, News, Prices & TipsAnthropic's Mythos AI Assists in Creating macOS Exploit in Five Days
Security researchers from a Palo Alto-based firm successfully developed a working macOS exploit in just five days, bypassing Apple's Memory Integrity Enforcement (MIE) security system. The breakthrough was achieved with significant assistance from Anthropic's Mythos Preview AI model. MIE, a hardware-assisted memory safety system introduced by Apple last year, is designed to prevent memory corruption exploits and is integrated into recent iPhone and MacBook models. The researchers utilized Mythos Preview to identify bugs belonging to known classes and aided in the collaborative development of the exploit. While the AI generalized attack strategies quickly, human expertise was crucial for autonomously bypassing the new mitigation technology. The team has prepared a 55-page technical report but will withhold its release until Apple distributes a fix. This incident highlights the potent combination of advanced AI models and human expert knowledge, raising concerns about how current security mitigations will withstand an era of AI-assisted cyberattacks, termed by the researchers as the first 'AI bugmageddon.'
SlashdotFirst Apple M5 Memory Exploit Discovered Using Anthropic AI Grants Root Access
Security researchers from the team Calif have discovered the first known privilege escalation exploit targeting Apple's M5 architecture. The vulnerability allows a standard user to gain root (administrator) access on macOS by bypassing the chip's Memory Integrity Enforcement (MIE) feature. Notably, the discovery was assisted by Anthropic's AI model, Claude Mythos, highlighting the growing role of artificial intelligence in both offensive and defensive security research. While Macs are rarely used as servers, limiting the immediate widespread impact, the exploit is considered concerning due to the ease with which users could be tricked into executing the malicious command. Once executed, the exploit grants full system control, making it difficult to detect and remove. This disclosure comes amidst a challenging period for cybersecurity, following significant vulnerabilities in Linux and Microsoft systems. As of now, the Calif team is the only group to have publicly disclosed this specific issue, providing Apple with an opportunity to patch the flaw before potential malicious actors widely exploit it.
Latest from Tom's HardwareSecurity Team Uses Anthropic Mythos to Build macOS Exploit in Five Days
Cybersecurity researchers from the firm Calif have demonstrated how Anthropic’s Mythos Preview AI tool was utilized to construct a working macOS kernel exploit in just five days. The target was Apple’s newest M5 silicon, which features Memory Integrity Enforcement (MIE), a hardware-level security mechanism introduced in September 2025 alongside the iPhone 17 and A19 chips to prevent memory corruption attacks. Despite Apple’s significant investment in this security feature, the researchers successfully chained two bugs and techniques to bypass MIE and achieve a root shell. The team described their work as a glimpse into future threats for hardware and software designed before the advent of advanced AI tools like Mythos. This incident highlights the potent capability of AI in surfacing unknown vulnerabilities and accelerating exploit development. In response to the disclosure, Apple is reportedly developing a fix to address the identified flaws. The report serves as a warning about the evolving landscape of cybersecurity where AI-assisted attacks can rapidly compromise systems previously considered secure.
Latest from TechRadarAnthropic AI Assisted in Breaching MacOS Security; Apple Investigating
Security researchers from Calif, a Palo Alto-based firm, successfully bypassed Apple's Memory Integrity Enforcement (MIE) protection on MacOS using an early version of Anthropic's Claude Mythos AI model. The team constructed the exploit in just five days by combining two system failures with memory corruption techniques, resulting in a privilege escalation vulnerability that could potentially grant full device control. Apple confirmed it is reviewing the detailed 55-page report submitted by the researchers, reiterating that security remains its top priority. While the AI significantly accelerated the attack's execution by reproducing known patterns, experts noted that human expertise was still essential for novel technique development. This incident highlights the growing capability of AI tools in vulnerability research and has intensified political debates regarding national security. The White House, previously resistant to Mythos's release due to defense concerns, is now reconsidering its regulatory approach to advanced AI models. The event underscores the dual-use nature of cybersecurity AI, as demonstrated by Anthropic's Project Glasswing initiative involving major tech companies.
ExameResearchers Use Anthropic Mythos AI to Exploit macOS Security Flaws
Researchers have successfully utilized Anthropic's Mythos AI tool to identify and exploit security vulnerabilities within Apple's macOS operating system, which is widely regarded as one of the most secure platforms globally. According to reports, the AI tool was able to link two minor software flaws, allowing it to corrupt the Mac's memory and access sensitive data. This exploit potentially grants hackers full control over a victim's computer. The discovery highlights the escalating threat posed by AI-assisted cyberattacks, where artificial intelligence can automate the identification of complex vulnerability chains that might remain hidden from traditional security audits. Apple is currently reviewing the findings provided by the researchers. This incident underscores the growing intersection of advanced AI capabilities and cybersecurity risks, demonstrating how tools like Mythos can challenge even robust security architectures. The report, citing The Wall Street Journal, emphasizes the need for enhanced security measures to counteract AI-driven exploitation techniques in modern operating systems.
Times of IndiaAnthropic's Mythos AI Identifies Security Flaws in Apple macOS
Anthropic's advanced AI model, Claude Mythos, has successfully assisted security researchers in identifying critical vulnerabilities within Apple's macOS operating system. According to reports, researchers from the Palo Alto-based firm Calif utilized the closed-access AI tool to discover a kernel memory corruption exploit affecting Apple's M5 chips. This exploit, described as the first of its kind publicly disclosed for this hardware, could potentially grant unprivileged local users complete control over affected devices. Anthropic has restricted public access to Mythos due to its potent capability in generalizing attack patterns across known bug classes, limiting its availability to select partners like Apple and specialized security firms. While Calif indicated that full technical details would remain confidential until Apple patches the vulnerabilities, recent macOS update notes suggest a fix submitted by Calif in collaboration with Anthropic has already been addressed. Apple reaffirmed its commitment to security, stating it takes such reports seriously. This development highlights the growing role of artificial intelligence in both discovering and mitigating complex software security threats within major technology ecosystems.
MashableCalif Team Uses Anthropic Mythos to Build macOS Exploit in Five Days
Security researchers from the Calif team have revealed how they utilized Anthropic’s Mythos Preview AI model to develop a working macOS kernel memory corruption exploit in just five days. This achievement successfully bypassed Apple’s Memory Integrity Enforcement (MIE), a hardware-assisted security system built on Arm’s Memory Tagging Extension (MTE) that Apple spent five years and billions of dollars developing. The exploit targets the new M5 silicon, marking the first public kernel memory corruption exploit on this architecture. The team, including researchers Bruce Dang, Dion Blazakis, and Josh Maine, discovered the vulnerabilities accidentally in late April 2026 and finalized the tooling by May 1st. The attack chain combines two bugs and several techniques to escalate privileges from an unprivileged local user to a root shell on macOS 26.4.1. While a detailed 55-page technical report exists, the team has withheld it until Apple releases a patch. This incident highlights the potential for advanced AI models to accelerate the identification of complex security vulnerabilities and challenge robust hardware-based defense mechanisms.
9to5MacAnthropic's Mythos AI Bypasses Apple macOS Security Systems
Security researchers have confirmed that Anthropic's unreleased AI model, Mythos, successfully hacked macOS by bypassing Apple's security systems in an unprecedented manner. Mythos, described as an early and more powerful version of the Claude AI software, demonstrated the ability to execute an escalation exploit. This exploit, if utilized maliciously, could allow hackers to gain full control over a Mac computer, overriding existing security measures. Due to the severity and novelty of the vulnerability, Anthropic engineers have deemed the model too dangerous for public release, citing its exceptional proficiency in identifying security exploits. The discovery was significant enough that the researchers, reportedly impressed by the AI's capabilities, traveled directly to Apple's headquarters in Cupertino to disclose their findings. The incident highlights growing concerns regarding the potential for advanced artificial intelligence to identify and leverage software vulnerabilities, posing new challenges for cybersecurity protocols in major tech ecosystems.
AppleInsider NewsAnthropic’s Mythos AI Helps Expose New macOS Security Vulnerabilities
A new report reveals that Anthropic’s secretive Mythos AI model was instrumental in uncovering significant security vulnerabilities in Apple’s macOS operating system. Security researchers from Calif, a Palo Alto-based firm, utilized an early version of Mythos in April to identify techniques that circumvent Apple’s advanced security protections. The resulting exploit is a privilege escalation attack that links two distinct bugs to corrupt system memory, potentially allowing hackers to gain unauthorized access to restricted parts of the device. Calif’s team compiled a 55-page report and delivered it directly to Apple’s headquarters in Cupertino for validation. While the AI model aided in discovery, Calif’s CEO Thai Dong emphasized that human cybersecurity expertise was essential to execute the attack. Apple has confirmed it is reviewing the findings and treating the potential vulnerabilities with high priority. Details of the specific bugs will remain confidential until Apple releases patches, which are expected to be deployed quickly. This incident highlights both the powerful capabilities of advanced AI models in cybersecurity research and the ongoing challenges tech giants face in securing their ecosystems against sophisticated, AI-assisted threats.
9to5Mac