Bank of America had to hire 1,000 contractors to patch software flaws spotted by Anthropic Mythos
THE SO WHAT
When an AI system forces a bank to hire 1,000 contractors and spend $100M+ to patch legacy flaws, the bottleneck isn’t detection—it’s remediation capacity. If you deploy AI code or security scanners at scale, budget for the downstream fix pipeline now or you’ll surface more risk than your org can realistically close.
READ THE SOURCE
MORE FROM THE WIRE
Applied AIFTC Subpoenas OpenAI and Anthropic Over AI Safety Concerns
Regulators are moving from speeches to subpoenas—AI safety incidents are now a consumer protection issue with discovery, not just PR. If you ship AI to end users, assume your incident logs, red-team reports, and mitigation workflows are future evidence and tighten them accordingly.
Applied AIMeta releases Muse Spark Open, saying it is the most capable US open-weight model, with 816B total and 42B active parameters and a 1M-token context window
A 42B active-parameter, 1M-token open-weight model tuned for agentic use on Apple desktops compresses the gap between local and cloud assistants. If you're building agents, you now have a credible on-device or hybrid path—revisit your dependency on single-vendor hosted LLMs.
Applied AIMeta Releases Open AI Model While Keeping Its Best Tech Closed
The split between downloadable models and closed, revenue-focused frontier systems is now explicit—"open" is a funnel, not the product. When you pick a model family, assume capability and licensing divergence over time and negotiate for upgrade and portability, not just current benchmarks.
Applied AIFriends don’t let friends feed their data to AI agents
Consumer AI assistants that ambiently capture bystanders’ data without clear consent are creating a new privacy externality—your users’ friends and coworkers become the de facto attack surface. If your product records or transcribes shared spaces, you need an explicit policy and UX for third-party consent this quarter, or regulators and app stores will eventually write it for you.