AI & MLAWS Sandbox Targets Rogue AI Agents
AWS's new Strands Box uses OS isolation and custom policies to curb AI agent behavior, but security experts warn controls have gaps.
> Search
19 results for “sandbox”
AI & ML
AI & MLAWS sandbox aims to tame rogue AI agents
Amazon's new open-source Strands Box uses OS-level isolation and policy rules to stop autonomous agents from going rogue.
SecurityDocker Fixes macOS Sandbox Escape Flaws
Two Docker Sandboxes vulnerabilities let malicious guest code read or modify macOS host files, with fixes shipped in version 0.42.0.
SecurityHomebrew 7.0.0 adds built-in vuln scanner
Homebrew 7.0.0 ships a native GUI, a Homebrew-specific advisory database, and stronger sandboxing as attackers keep targeting the package manager.
Cyber CrimeTencent Patch Leaves Sogou Browser Exposed
A one-click flaw in Sogou Input Method allowed UNC3569 to deploy GrayRabbit malware, but the underlying Chromium engine remains outdated and unsandboxed.
AI & MLChatGPT's Hidden Data Channel Raises Privacy Stakes
A planted prompt could make ChatGPT silently exfiltrate Gmail data via a covert channel between sandboxes.
AI & MLAI agents colluded, shared answers on public wiki
A swarm of 3,700 OpenAI agents made 18,000 wiki posts, discussing sandbox escapes and sharing test answers.
SecurityServiceNow Tackles Maximum-Severity Flaws
Patches cover three critical code injection bugs and a sandbox escape in the Now Platform.
SecurityNCSC's Agentic AI Controls Raise the Stakes for Autonomy
UK NCSC issues interim guidance urging sandboxing, human oversight, and access limits for agentic AI systems.
SecurityAI Labs Compete Over Rogue Agent Claims
Anthropic and OpenAI trade narratives of accidental sandbox escapes, raising questions about the safety of their frontier models.
SecurityCursor Flaw Executes Code Before Trust
A Cursor bug let repositories run commands pre-trust, even with the sandbox enabled.
SecurityCritical IBM Langflow Sandbox Flaw Found
A critical input validation vulnerability in IBM Langflow OSS allows for potential system compromise, requiring immediate attention from administrators.
SecuritySharedRoot Sandbox Escape in Claude Cowork
A critical vulnerability in Anthropic's Claude Cowork allows AI agents to escape their Linux virtual environment and access host macOS data.
ServiceNow Sandbox Breach Escalates
Enterprise AI systems face active exploitation of a high-complexity remote code execution flaw following a recent patch release.
SecurityCISA Adds FortiSandbox Flaw to KEV List
Federal agencies must address a critical OS command injection vulnerability in Fortinet products by July 19, 2026.
SecurityCISA Warns of Active FortiSandbox Flaw
Federal agencies must address an actively exploited OS command injection vulnerability in Fortinet products by July 19, 2026.
SecurityTwig Template Engine Vulnerability Allows Sandbox Policy Bypass
A critical vulnerability in Twig versions 3.9.0 through 3.25.0 allows sandboxed templates to bypass security policy enforcement.
SecurityCISA Adds Actively Exploited Fortinet FortiSandbox Flaw to KEV Catalog
CISA has confirmed active exploitation of an OS command injection vulnerability in Fortinet FortiSandbox, mandating urgent remediation for federal agencies.
SecurityWhatsApp Exploits Can Now Breach Host Systems via OpenClaw
Three vulnerabilities in the OpenClaw AI assistant allow attackers to bypass sandbox restrictions and execute code on host environments.