AMD is making a hard bet on the future of inference. The chipmaker is acquiring Toronto startup Taalas, which hardwires specific AI models directly into silicon rather than running them on general-purpose hardware. It is a clear signal that AMD sees specialized, cost-efficient inference chips as the next battleground, especially as DeepSeek's V4 Flash model just achieved an 89% score on the ARC-AGI benchmark for just two cents per task. Efficiency gains like those are precisely what makes purpose-built silicon compelling. As models get smarter, the push to govern them intensifies. AI music generator Suno will begin watermarking all AI-generated songs and banning unauthorized voice cloning, a policy shift driven by legal pressure and mounting concerns about deceptive audio. Meanwhile, a former US National Cyber Director is echoing Asimov's robot laws, warning that the biggest risk is autonomous AI choosing its own rules. Even law enforcement is grappling with transparency: ICE will equip officers with body cameras but has reserved the right to release footage only when it serves the agency’s best interests, raising familiar questions about accountability. Cybersecurity remains a tinderbox in both consumer and critical infrastructure. Framework Computer notified all customers of a data breach after a compromised Metabase BI service was exploited, a stark reminder that even transparent hardware companies can be undone by third-party tools. More alarmingly, water system controllers in at least twelve US states have been hacked, likely by Iran, prompting an ex-NSA chief to demand that critical infrastructure never touch the public internet. On a local level, New Orleans is now using AI to answer 911 calls, pushing automation into one of the most high-stakes public services imaginable.