Anthropic’s $200-per-month Claude Max 20x plan is facing a proposed class action alleging that its headline multiplier applies only to a five-hour session, not to the weekly capacity customers actually use. Based on Anthropic’s own published ranges, plaintiff Karl Kahn says Max 20x delivers roughly six to eight times the Pro plan’s weekly usage, while Max 5x delivers about 3.5 times—not the advertised fivefold and 20-fold increases.

Source

Tim Cook ends his 15-year tenure as Apple CEO on August 31, 2026, handing control to hardware engineering chief John Ternus on September 1. Cook will remain executive chairman, where he is expected to retain a major role in Apple’s government and policy relationships.

Source

Cloudflare has launched Adaptive Intelligence, a Bot Management engine that continuously retrains on live traffic and changes its detection logic as attackers adapt. The system is designed not merely to block bots, but to make repeated automation attempts slower, less predictable, and increasingly expensive to operate.

Source

Anthropic has expanded its Compliance API to expose Claude Code’s local sessions, including prompts, shell commands, file operations, and MCP activity. The update gives enterprise security teams a stronger audit trail for endpoint-based AI agents, but the transcripts still cannot determine whether an action was authorized or appropriate.

Source

Windows Autopilot Device Preparation is gaining a device-association workflow that can identify a physical Windows 11 device through TPM-based attestation before enrollment, removing the need to import hardware hashes or enter a corporate identifier during OOBE. The change strengthens the case for making Device Preparation the default in cloud-native Intune environments, while original Autopilot remains necessary for pre-provisioning, self-deploying devices, and hybrid join.

Source

OpenAI has begun offering some large customers a new deal: they pay only when its AI completes an assigned task, such as resolving a support request. The trial moves beyond subscriptions and usage fees toward outcome-based pricing, a model OpenAI is also considering for commercially valuable discoveries and other enterprise results.

Source

ChatGPT has become the first AI chatbot designated a Very Large Online Search Engine under the EU Digital Services Act (DSA), giving regulators expanded powers to examine how it operates. OpenAI now has four months to address systemic risks tied to illegal content, child safety, election interference, and other effects on European users.

Source

A PowerShell script can now compare Microsoft Graph’s current permissions with a stored baseline and alert administrators when Microsoft adds narrower or new access options. The automation also uncovered a problem in Azure Automation: Microsoft Graph Files cmdlets were unavailable in the PowerShell 7.4 runtime, forcing a switch to direct Graph API requests.

Source

Microsoft has confirmed that Windows 11’s optional KB5120998 preview update resets mouse personalization settings, validating reports of custom cursors turning white and wallpapers reverting. The issue can replace high-DPI pointers with oversized white cursors, revert animations, and prevent users from restoring their previous settings.

Source

Broadcom has now moved beyond the private AI and agent-security agenda previewed for VMware Explore 2026, announcing the VMware AI Factory, AgentMinder, and new TrueSource open-source security services. The package positions VMware Cloud Foundation as a turnkey platform for running production AI on enterprise infrastructure rather than sending sensitive workloads to public clouds.

Source

General Motors’ rollout of passkeys to roughly 200,000 employees shows that the difficult part is not enabling FIDO2 authentication—it is replacing every weak-method dependency around it. With Microsoft Entra ID set to make passkeys the default on September 1, 2026, and retire Microsoft-provided SMS and voice MFA on February 1, 2027, GM’s experience offers administrators a practical migration plan for onboarding, guests, factory workers, unmanaged devices, and VDI users.

Source

Linux 7.3-rc1 has become one of the kernel’s largest first release candidates, reaching nearly 41 million lines of source code. AMD’s new GPU definitions and driver support account for about one-third of the merge-window patch, making the release especially relevant for teams testing upcoming AMD hardware.

Source

OpenClaw 2.0 is the project’s biggest update yet, but its attempt to regain relevance comes as competing AI products have absorbed many of its original advantages. The release rebuilds installation, the browser interface, collaboration, memory, and release tooling after an unusually long seven-week pause, yet its more cautious feature set may not be enough to restart the platform’s momentum.

Source

Debian developers have rejected a blanket ban on large language model (LLM)-assisted contributions, while leaving disclosure of AI use optional. The decision preserves existing human review and makes contributors responsible for checking quality, security, and licensing before submitting work.

Source

Anthropic is revoking Claude login sessions after infostealer malware began hijacking authenticated browser sessions and consuming users’ usage limits. Because the stolen sessions are already trusted, attackers can access accounts without passwords or two-factor authentication.

Source

Z.ai’s GLM-5.3-Flash, previously tested anonymously as Ox Alpha on OpenRouter, combines near-frontier coding and agentic performance with open weights, local deployment options, and exceptionally low inference costs. Matthew Berman highlights the model’s efficiency—and the broader significance of its reported ability to serve massive workloads on Chinese AI accelerators.

Source

Anthropic is changing how much Claude Code usage subscribers receive, but its policy is difficult to interpret. The service uses weekly and five-hour limits rather than a simple prompt count. Weekly usage depends on factors such as conversation length, files, tool calls, model choice, and reasoning effort, so Anthropic does not provide a clear conversion to messages, tokens, or completed coding tasks.

Source

Microsoft has confirmed that Windows 11’s “Turn on virus protection” warnings are false positives caused by a recent Microsoft Defender Antivirus update. Defender remains active, but the alerts can keep returning until a future update fixes the problem.

Source

Popular AI coding agents are poor at estimating both how long a programming task will take and how long they have already been working, according to tests of Anthropic’s Claude Code and OpenAI’s Codex. That makes them unreliable for long-running jobs unless administrators give them an external timer and explicit time limits.

Source