xAI completed its rebrand to SpaceXAI on July 7, folding Musk's AI lab, SpaceX, and the recently acquired Cursor into a single corporate entity announced on X.

• Strategic rationale: the merger targets eventual space-based compute infrastructure, with declining rocket launch costs making orbital data centers increasingly plausible.
• Cursor absorbed: the IDE acquisition now operates under SpaceXAI, meaning enterprise dev tool contracts route through the same entity managing Grok model development.

BeyondTrust released emergency security updates on July 7 addressing multiple critical authentication bypass vulnerabilities in its Remote Support and Privileged Remote Access products.

• Severity: flaws carry critical ratings and could allow unauthenticated attackers to take full control of enterprise remote access sessions, per the advisory.
• Patch urgency: BeyondTrust Remote Support and PRA are deployed widely as privileged access gateways, making unpatched instances a direct path to lateral movement across enterprise networks.

OpenAI shipped a major update to Codex Mobile for iOS on July 7, enabling users to create, search, open, fork, and manage Codex agent tasks directly from any conversation.

• Capability shift: the update moves Codex Mobile from a task-viewer into a full task-management surface, letting developers kick off and redirect autonomous coding runs from a phone.
• Context: this ships alongside confirmed plans to embed GPT-5.6 Sol Ultra in the Codex client, positioning the mobile app as the control plane for the highest-compute reasoning tier.

Zscaler published research on July 7 showing that multiple frontier AI agents, including expensive enterprise-grade models, fall victim to indirect prompt injection attacks that most humans would recognize as fraud.

• Finding: agents trained to treat structured metadata as high-signal instructions cannot separate untrusted web content from trusted directives when both share the context window, an architectural gap not a behavioral one.
• Implication: Zscaler's researchers argue defenses must be architectural, meaning network-layer interception and sandboxing, not just model-level guardrails, a direct pitch for zero-trust proxy controls around agent traffic.

CISA's Attack Surface Evaluation team is actively using Anthropic's Mythos model to scan federal code repositories for exploitable vulnerabilities, according to three sources cited by Reuters.

• Scope: the effort targets government code repositories for bugs exploitable by foreign intelligence services or cybercriminals, making it one of the first confirmed large-scale sovereign AI security deployments.
• Timing tension: the deployment runs alongside an emergency export control directive that restricted Mythos access to foreign nationals, implying domestic government use was explicitly carved out.