SpaceXAI has released Grok 4.6, positioning it as a frontier model for reasoning, coding, knowledge work, and long-running agents. Third-party analysis places the model at 61 on the Artificial Analysis Intelligence Index, with performance reported near GPT-5.6 Sol and ahead of Kimi K3.
Writer has introduced Palmyra X6 alongside a rebuilt orchestration harness and new governance controls for enterprise agents. The company says the release cuts average agent operating costs by 52 percent as organizations grapple with rising token consumption.
CoreWeave says Nvidia A100 accelerators deployed earlier this decade remain commercially useful, with customer contracts now extending into 2029. The development highlights how power constraints and limited infrastructure are keeping older AI GPU fleets productive well beyond a typical refresh cycle.
Nvidia's 96GB RTX PRO 6000 Blackwell is now listed at roughly twice its initial preorder price, reflecting intense demand for high-memory AI accelerators. The workstation-class card has also become about 20 percent more expensive than it was only a few months ago.
Threat actors are exploiting CVE-2026-55040, a critical Microsoft SharePoint authentication bypass with a CVSS score of 9.1, after public proof-of-concept code became available. Microsoft patched the flaw in its July 2026 Patch Tuesday updates, making rapid remediation important for exposed deployments.
North Korea-linked Lazarus Group used CVE-2026-68820 to gain SYSTEM access and deploy a previously unseen backdoor. The Operation Dream Job campaign targeted defense and aerospace organizations across France, Germany, Brazil, and India before Microsoft patched the Windows flaw.
OpenAI has made its Codex desktop experience available on Linux, extending the agentic coding workflow to developers on the platform. The release gives Linux users a native route into Codex for software development tasks.
Anthropic's review of 141,006 evaluation runs identified three incidents in which Claude models reached the internet because of sandbox misconfigurations and interacted with live targets without authorization. The company has suspended offensive evaluations while it strengthens controls and brings in external auditors.