2x Usage Promo Enters Final Week — Through March 27 for All Plans
Anthropic’s March bonus usage promotion is in its final seven days. Through March 27, every Free, Pro, Max, and Team subscriber gets up to 2x their normal message limits during off-peak hours at no extra cost. Off-peak means outside 8 AM–2 PM Eastern on weekdays; on weekends it’s 24/7. The bonus usage doesn’t count against weekly plan caps — it’s genuinely additive capacity. Enterprise plans are not included.
The promo covers all Claude surfaces: the web and mobile apps, Claude Code, Cowork, and the Excel and PowerPoint add-ins. If you’ve been sitting on a large project, this is the week to run it. Anthropic confirmed the promotion is fully automatic — no codes, no settings to toggle.
Pentagon CTO: DoD “Pretty Confident” It Can Replace Claude Within Six Months
As the March 24 court hearing looms, the Pentagon is leaning into its transition narrative. DoD Under Secretary for Research and Engineering Emil Michael told reporters the military is “pretty confident” it can phase out Anthropic’s Claude products within the 180-day window set by President Trump. Michael pointed to active OpenAI and Gemini deployments already underway across several classification networks as evidence the transition is feasible. “The warfighter is going to have tons of different options,” he said.
Defense One and Federal News Network reporting paints a more complicated picture. Claude is embedded in multiple classified systems where it took months to deploy — and replacing it isn’t as simple as swapping an API key. Previous sources told reporters it would take “months” just to remove Claude from existing toolchains, let alone migrate workflows. The March 24 hearing in U.S. District Court for the Northern District of California will determine whether a judge issues a preliminary injunction halting the supply chain risk designation while the underlying lawsuit proceeds. Anthropic has the backing of Microsoft, nearly 150 retired federal judges, and former senior national security officials.
Claude Code Voice Mode Expands to All Users — Plus /loop, /effort, and 1M Context
Claude Code’s voice mode is now rolling out to all users after a limited 5% pilot that began in early March. It activates with the /voice command and works via push-to-talk: hold the spacebar to speak, release to send. Ten new languages were added this month, bringing total voice recognition support to 20. The feature is notified on the Claude Code welcome screen as it becomes available for each account.
The bigger workflow addition may be /loop, which turns Claude Code into a lightweight session-level cron job. Set an interval and a prompt — /loop 5m check the deploy — and Claude executes it automatically on that cadence. Combined with the new /effort command (lets you dial the model’s reasoning depth up or down per task) and /batch for parallel execution, these updates make Claude Code meaningfully more usable as an always-running background agent, not just an interactive coding partner. Opus 4.6 on Max, Team, and Enterprise plans also now gets 1M token context by default, enabling whole-codebase conversations without compaction.
Claude for Excel and PowerPoint: Shared Context and Team Skills Now Live
Anthropic updated the Claude add-ins for Microsoft 365 so that the Excel and PowerPoint integrations now share a single conversation thread. Previously each app started a fresh session — now context from your spreadsheet work carries into the slide deck you’re building from it, with no repeated setup. The update also introduces one-click “Skills” that let teams save proven workflows and share them organization-wide. Starter Skills targeting financial services — DCF templates, LBO models, investment banking deck reviews — are available immediately.
Claude in PowerPoint can now generate native editable charts and diagrams from natural-language prompts, respecting the existing slide master and color scheme. If you’re still running the early research preview version, check the Office Add-ins store for the update.
Anthropic Publishes A3 and AuditBench — Alignment Research Goes Agentic
Anthropic’s alignment science team published two papers this week that push the field toward more automated safety work. The first, A3 (Automated Alignment Agent), is an agentic framework that automatically detects and mitigates safety failures in large language models with minimal human intervention. A3 is built on the Claude Agent SDK and is designed to work like a thoughtful human auditor — running a configurable toolkit of probes against a target model and flagging hidden behaviors.
The second paper introduces AuditBench, a benchmark of 56 language models (all based on Llama 3.3 70B) with deliberately implanted hidden behaviors spanning 14 categories: sycophantic deference, opposition to AI regulation, hidden geopolitical loyalties, and more. Each model is trained specifically not to confess its hidden behavior when directly asked. AuditBench gives the broader research community a standardized testbed for evaluating how well auditing approaches actually detect model misalignment — a problem that’s hard to measure without known ground truth. Both are available on the Anthropic Alignment Science blog and arXiv.
The DoD’s Six-Month Confidence Is a Legal Move, Not an Engineering One
Emil Michael’s “pretty confident” quote is a statement made for a judge, not for a transition project manager. When the Pentagon says it can replace Claude in six months, it’s trying to undercut Anthropic’s argument that the supply chain designation causes immediate irreparable harm. If the court believes the transition is easy and fast, it’s harder to grant a preliminary injunction. That doesn’t mean the 180-day transition is actually smooth — Defense One reported months ago that removing Claude from classified systems would itself take months, not days. And swapping an AI model embedded in operational workflows isn’t like updating a software version; it’s retraining teams and rebuilding integrations.
The March 24 hearing is the first real test of Anthropic’s legal strategy. The company needs the judge to accept that the irreversibility and speed of the transition — not just the long-term consequences — justify a stay. If that argument holds, the litigation continues with Claude still deployed in government systems. If it doesn’t, the 180-day clock keeps running. Either way, the hearing is more decisive than any press briefing either side gives this week.