XBOW used GPT-5.6 Sol to build a $250,000 Chrome exploit in six days
THE SO WHAT
Offense timelines just collapsed from quarters to days when paired with frontier models—your patching, red-teaming, and disclosure processes have to assume exploit development is now near-real-time. If four experts needed 6–9 months pre-LLMs and one team now ships a Chrome full chain in six days, security moves from perimeter tools to continuous, AI-augmented adversarial testing.
READ THE SOURCE
MORE FROM THE WIRE
Applied AINotorious 3D-Printed Gun Activist Debuts ‘AI That Can Generate Guns’
Open-source CAD plus generative models collapses the skill barrier for weapons design—governance will shift from controlling files to controlling capabilities and distribution. If you operate in any dual-use domain, assume your design tools are now part of the regulatory conversation.
Applied AIDeepSeek Pushes Out Updated AI Model It Says Is Faster, Leaner
DeepSeek pushing a faster, leaner model is another data point that cost-per-token and latency—not just raw IQ—are the competitive front. If you’re standardizing on a single US lab, budget time this quarter to benchmark at least one Chinese or open alternative on your real workloads.
Applied AIAs AI giants rack up potential legal liabilities of tens of billions amid safety incidents, insurers worry claims may outpace limits on existing coverage
If insurers believe AI-related claims can outrun current coverage limits, your real constraint becomes risk capital, not just compute or talent. Expect tighter exclusions, higher premiums, and more intrusive underwriting—operators need a clean risk story, logged controls, and clear incident playbooks or they will pay for the uncertainty in cash and deal friction.
Applied AIEU investigates OpenAI and Anthropic over agent hacks, Euronews reports
The EU AI Office is moving from abstract rules to concrete RFIs on security, external evals, and agent misuse—regulators are treating agentic behavior as a live risk surface. If you deploy or build on these models in Europe, assume you’ll need auditable monitoring of how agents act in the wild, not just model cards and DPIAs.