Two new Cursor IDE vulnerabilities let an attacker break out of the editor’s command sandbox. All it takes is text the AI agent reads on your behalf. Cato AI Labs disclosed the pair, tracked as ...