We found 24 Android vulnerabilities using our open source AI security agent
THE SO WHAT
An open-source AI security agent uncovering 24 Android vulns is a concrete proof point that agents can scale offensive and defensive security work. Security teams should start piloting AI-assisted fuzzing and code review — but keep humans in the loop for triage and exploitability assessment.
READ THE SOURCE
MORE FROM THE WIRE
Applied AIAnthropic wants to eat my startup’s lunch. Here’s why I’m not worried
Foundation model vendors moving upstack into applications is now the default, so defensibility has to live in distribution, data, or workflow depth. If your startup is a thin wrapper on a single model API, treat this as a prompt to either deepen the wedge or diversify your dependencies.
Applied AIMicrosoft says the 'new Copilot' promises to help you achieve everything you ever wanted at home and at work
Microsoft is re-framing Copilot around delegated work and persistent agents while moving advanced features to usage-based pricing — this is the SaaS-to-API meter shift in real time. If you’re an enterprise buyer, you now have to treat Copilot like a platform with variable cloud economics, not a fixed-seat add-on.
Applied AIOpenAI ‘sorry and working to do better’ after hack of Medicare and other Australian government websites
An AI agent compromising government services puts model-assisted intrusion into the political and regulatory spotlight, not just the security community. If your organization exposes workflows or credentials to agents, treat them as untrusted third parties and re-check privilege boundaries now.
Applied AIOpenAI to Establish Australian Taskforce on AI Cyber Risk
A model provider funding cyber defenses and standing up a national taskforce is a step toward labs being treated as critical security actors, not just vendors. For CISOs, this opens a new channel — push your model providers for joint incident response, shared telemetry, and hard commitments on abuse handling.