Sandbox Escape | Breaking Cybersecurity News | The Hacker News

Cybersecurity researchers have uncovered a sandbox escape vulnerability in Anthropic's Claude Cowork that makes it possible to break out of the confines of a Linux virtual machine (VM) within which the agent runs to read or write files anywhere on the Mac. Accomplish AI, which shared details of the vulnerability with The Hacker News ahead of publication, said about 500,000 macOS users running local Cowork sessions were affected prior to it being patched. It has been codenamed SharedRoot . "We connected a folder to a fresh Claude Cowork session, sent one short message, and watched the agent escape the sandbox," Oren Yomtov, principal security researcher at Accomplish AI, said . "From inside the VM, it reached the host Mac and read and wrote files all over it, far outside the folder we'd connected, with no permission prompt anywhere." With this level of access, the agent can access any data stored on the Mac via the user's account, including SSH ke...

Details have emerged about three now-patched security flaws in the OpenClaw personal artificial intelligence (AI) assistant that, if successfully exploited, could enable credential theft, privilege escalation, and arbitrary code execution on the host. A brief description of the high-severity vulnerabilities is as follows - GHSA-hjr6-g723-hmfm (CVSS score: 8.8) - An operating system command injection and an incomplete list of disallowed inputs vulnerability impacting the host execution environment filtering mechanism that could allow for executing or persist actions beyond the caller's intended authorization. GHSA-9969-8g9h-rxwm (CVSS score: 8.8) - An operating system command injection and an incomplete list of disallowed inputs vulnerability impacting the host execution environment filtering mechanism that could allow for executing or persist actions beyond the caller's intended authorization. GHSA-575v-8hfq-m3mc (CVSS score: 8.4) - A path traversal and link f...

Two flaws in Cursor, an AI code editor, could let a single, ordinary-looking prompt break out of the editor's safety sandbox and run any command on a developer's computer. There is no click to fall for and no approval box to ignore. Cato AI Labs found the pair and named them DuneSlide . They are tracked as CVE-2026-50548 and CVE-2026-50549, both rated 9.8 out of 10 (or 9.3 under the newer CVSS 4.0 scale). The fix is already out. Both bugs are patched in Cursor 3.0, released April 2, and every version before 3.0 is affected. Cursor's maker says more than half the Fortune 500 use the tool, so if you run it, update now. What the sandbox was for, and how it broke Starting in the 2.x line, Cursor runs the terminal commands its AI agent issues inside a sandbox by default: a locked box that limits what those commands can touch, so a stray instruction cannot wreck the machine. DuneSlide is about getting out of that box. The way in is prompt injection . The attacke...

Two things landed within days of each other this week. A security startup reported 21 previously unknown vulnerabilities in FFmpeg, the media library inside almost everything that touches video, all of them found by an autonomous AI agent. The same week, Google shipped Chrome 149 with patches for 429 security bugs, the most ever in a single release. Only the FFmpeg bugs were found by AI. Chrome's record landed after Google overhauled its bounty program to cope with a flood of AI-generated reports. The mechanisms differ, but the pressure is the same: AI is putting more vulnerabilities in front of the people who have to deal with them, and faster than before. The FFmpeg findings come from depthfirst , whose autonomous security agent scanned the project's roughly 1.5 million lines of C and produced 21 confirmed zero-days, each with a reproducible proof-of-concept input. The company puts the cost of the run at around $1,000. Several of the bugs had been latent for 15 to 2...

Cybersecurity researchers have disclosed a set of four security flaws in OpenClaw that could be chained to achieve data theft, privilege escalation, and persistence. The vulnerabilities, collectively dubbed Claw Chain by Cyera, can permit an attacker to establish a foothold, expose sensitive data, and plant backdoors. A brief description of the flaws is below - CVE-2026-44112 (CVSS score: 9.6/6.3) - A time-of-check/time-of-use (TOCTOU) race condition vulnerability in the OpenShell managed sandbox backend that allows attackers to bypass sandbox restrictions and redirect writes outside the intended mount root.  CVE-2026-44113 (CVSS score: 7.7/6.3) - A TOCTOU race condition vulnerability in OpenShell that allows attackers to bypass sandbox restrictions and read files outside the intended mount root. CVE-2026-44115 (CVSS score: 8.8) - An incomplete list of disallowed inputs vuln...

Google on Tuesday rolled out fixes for six security issues in its Chrome web browser, including one that it said has been exploited in the wild. The high-severity vulnerability in question is CVE-2025-6558 (CVSS score: 8.8), which has been described as an incorrect validation of untrusted input in the browser's ANGLE and GPU components. "Insufficient validation of untrusted input in ANGLE and GPU in Google Chrome prior to 138.0.7204.157 allowed a remote attacker to potentially perform a sandbox escape via a crafted HTML page," according to the description of the flaw from the NIST's National Vulnerability Database (NVD). ANGLE, short for "Almost Native Graphics Layer Engine," acts as a translation layer between Chrome’s rendering engine and device-specific graphics drivers. Vulnerabilities in the module can let attackers escape Chrome’s sandbox by abusing low-level GPU operations that browsers usually keep isolated, making this a rare but powerful path...

Google has rolled out security updates to fix seven security issues in its Chrome browser, including a zero-day that has come under active exploitation in the wild. Tracked as  CVE-2023-6345 , the high-severity vulnerability has been described as an integer overflow bug in Skia, an open source 2D graphics library. Benoît Sevens and Clément Lecigne of Google's Threat Analysis Group (TAG) have been credited with discovering and reporting the flaw on November 24, 2023. As is typically the case, the search giant  acknowledged  that "an exploit for CVE-2023-6345 exists in the wild," but stopped short of sharing additional information surrounding the nature of attacks and the threat actors that may be weaponizing it in real-world attacks. It's worth noting that Google released patches for a similar integer overflow flaw in the same component ( CVE-2023-2136 ) in April 2023 that had also come under active exploitation as a zero-day, raising the possibility that CVE-202...

Cybersecurity researchers have uncovered a new cloud targeting, peer-to-peer (P2P) worm called  P2PInfect  that targets vulnerable Redis instances for follow-on exploitation. "P2PInfect exploits Redis servers running on both Linux and Windows Operating Systems making it more scalable and potent than other worms," Palo Alto Networks Unit 42 researchers William Gamazo and Nathaniel Quist  said . "This worm is also written in Rust, a highly scalable and cloud-friendly programming language." It's estimated that as many as 934 unique Redis systems may be vulnerable to the threat. The first known instance of P2PInfect was detected on July 11, 2023. A notable characteristic of the worm is its ability to infects vulnerable Redis instances by exploiting a critical Lua sandbox escape vulnerability,  CVE-2022-0543  (CVSS score: 10.0), which has been previously exploited to deliver multiple  malware families  such as  Muhstik ,  Redigo , and  H...

Muhstik, a botnet infamous for propagating via web application exploits, has been observed targeting Redis servers using a recently disclosed vulnerability in the database system. The vulnerability relates to  CVE-2022-0543 , a  Lua sandbox escape flaw  in the open-source, in-memory, key-value data store that could be abused to achieve remote code execution on the underlying machine. The vulnerability is rated 10 out of 10 for severity. "Due to a packaging issue, a remote attacker with the ability to execute arbitrary Lua scripts could possibly escape the Lua sandbox and execute arbitrary code on the host," Ubuntu noted in an advisory released last month. According to  telemetry data  gathered by Juniper Threat Labs, the attacks leveraging the new flaw are said to have commenced on March 11, 2022, leading to the retrieval of a malicious shell script ("russia.sh") from a remote server, which is then utilized to fetch and execute the botnet binaries from anoth...