OpenAI's GPT-5.6 preview introduces Sol as the flagship model, with Terra for efficiency and Luna for lower-cost, faster workloads. The rollout is limited to selected partners, so teams watching API availability should treat this as a capability preview rather than a broad production launch.
Anthropic's Mythos 5 is reportedly back for a selected group of U.S. companies and agencies after negotiations with the U.S. government. Public access remains constrained, which keeps frontier-model planning tied to eligibility, compliance and regional availability.
TechCrunch highlighted a growing move by major AI and compute buyers toward in-house or partner-built inference silicon. OpenAI's reported Broadcom-backed Jalapeno chip sits alongside efforts from Google, Apple and SpaceX, signaling that AI infrastructure buyers want more leverage over cost, capacity and supply risk.
Tom's Hardware reports that Intel's flagship Nova Lake processor could use a 474W PL2 limit and require high-end LGA1954 motherboard designs with up to three 8-pin power connectors. If accurate, enthusiast and workstation boards will need more robust power delivery for the next high-core-count desktop generation.
The Kubernetes project says AI-assisted coding is increasing contribution volume, but maintenance and review capacity have not scaled at the same pace. The post frames AI code as useful when it lands inside strong ownership, testing and review practices, which matters for teams consuming cloud-native projects.
IBM and Red Hat are bringing Deloitte into an initiative focused on remediating security issues in open-source software. The partnership reinforces a trend toward coordinated vendor-backed maintenance for dependencies that sit underneath enterprise cloud platforms.
CISA ordered U.S. federal agencies to patch an actively exploited Cisco Unified Communications Manager Server vulnerability by Sunday. The short deadline signals real-world exploitation risk and makes Cisco voice and collaboration infrastructure a near-term patch priority.
Polymarket says hackers injected malicious frontend script through a compromised third-party vendor, leading to about $3 million in customer losses. The incident is another reminder that web supply-chain controls, vendor access limits and frontend integrity checks are part of production security, not just compliance.
A Show HN project from WorkWeave routes coding-agent requests to different models based on the task. The pitch is cost and quality control for AI-heavy development teams that already use Claude Code, Codex or Cursor and want routing without changing their workflow.
TechCrunch covered a dispute in which Papermark accused a startup of copying its open-source product, while the startup denied wrongdoing. The case is relevant for engineering teams because AI-assisted implementation can blur assumptions around licenses, attribution and clean-room development.