Skip to content
DailyPrompt

your daily dose of claude

01 announcement

Google to Plough Up to $40 Billion More Into Anthropic in Cash and Compute

Alphabet has committed up to $40 billion in fresh funding to Anthropic, structured as $10 billion now at a reported $350 billion valuation and another $30 billion if performance targets are hit. Alongside the cash, Google Cloud will provide five gigawatts of compute over the next five years, building on the trio's recent TPU deal with Broadcom. The move comes days after Amazon's $25 billion top-up and underscores the scramble among hyperscalers to lock in Claude as a flagship workload.

TechCrunch →
product

Anthropic Pins Claude Code's Monthlong Slump on Three Engineering Missteps

After weeks of user complaints, Anthropic confirmed that three separate changes between March and April quietly degraded Claude Code: a reasoning-effort downgrade from high to medium, a caching bug that wiped thinking history every turn, and a verbosity prompt that capped responses at 25 words between tool calls. All three were resolved by version 2.1.116 on 20 April, and the firm has reset usage limits for every paying subscriber as compensation. The company conceded that prioritising lower latency over reasoning quality was the wrong trade-off.

Fortune
A
Anthropic
@AnthropicAI

New Anthropic research: Project Deal. We created a marketplace for employees in our San Francisco office, with one big twist. We tasked Claude with buying, selling and negotiating on our colleagues' behalf.

12h View on X →
announcement

$ cat story_03.md

Anthropic Beefs Up Claude's Election Safeguards Ahead of US Midterms

Anthropic has rolled out an updated election integrity stack covering the 2026 US midterms and other global votes, combining new classifiers, threat-intelligence monitoring, and constitutional training. Claude Opus 4.7 and Sonnet 4.6 scored 95% and 96% respectively on a fresh political-balance benchmark, while a 600-prompt safety eval saw both models respond appropriately to election-related requests at least 99.8% of the time. Voters asking about registration or ballots are now routed to TurboVote via in-product banners.

$ open anthropic →
research No. 04

Project Deal: Anthropic Lets Claude Run a Real Marketplace for Its Own Employees

In a new research experiment dubbed Project Deal, Anthropic gave 69 San Francisco staff a $100 budget each and let Claude agents handle every buy, sell, and negotiation across an internal classifieds marketplace. Across four parallel runs, agents struck 186 deals worth more than $4,000, with stronger models like Opus 4.5 securing measurably better outcomes than Haiku 4.5 — though participants paired with weaker agents barely noticed the gap. The team says the findings hint at how agent-to-agent commerce could scale, and where policy will need to catch up.

Source: Anthropic →
A
Anthropic
@AnthropicAI

Markets of AI agents could provide value, but there are plenty of rough edges. Access to higher-quality models conferred a real advantage—and participants didn't notice. There are plenty of other ways things can go wrong. Policy and legal frameworks will need to adapt to keep up.

12h View on X →
05
product

Freshfields Rolls Claude Out to 5,700 Lawyers Across 33 Global Offices

Magic Circle firm Freshfields Bruckhaus Deringer has signed a multi-year deal to deploy Claude across its entire global workforce via the firm's proprietary AI platform, with usage already up roughly 500% in the first six weeks. The two firms will co-develop legal-focused workflows for research, contract review, due diligence, and multi-step agentic tasks, and Freshfields has secured early access to forthcoming Anthropic models. It is one of the largest Claude deployments inside the legal industry to date.

Law.com →
opinion

The Register: Anthropic Quietly 'Dumbed Down' Claude With Its Own Upgrades

Reading Anthropic's postmortem with a sharper eye, The Register argues that the firm's recent string of well-intentioned tweaks effectively shipped a worse Claude to paying users for weeks, even as the company publicly insisted nothing had changed. The piece highlights how a verbosity-cutting prompt, intended to make answers feel snappier, dragged coding eval scores down by three percent on Opus 4.6 and 4.7. It calls for far louder change-logs whenever harness or system-prompt tweaks land in production.

Read on The Register →
Breaking

Claude Code Gets a Cross-Platform Refresh Spanning Web, Mobile, and CLI

Anthropic has shipped a meaningful update to Claude Code's web and mobile experiences, with a redesigned UI, performance and reliability gains, and proper continuity between surfaces. Developers can now start a session on the web or phone and use the new 'claude --teleport' command to pick up exactly where they left off in their terminal, with full history and the right branch checked out. The refresh lands the same week Anthropic's postmortem closed the book on the model's recent quality wobble.

ClaudeDevs (X)
C
ClaudeDevs
@ClaudeDevs

We've refreshed Claude Code on the web and mobile. A few things that recently shipped 🧵

9h View on X →
A
Amanda Askell
@AmandaAskell

It's odd to be living through what feels like one of the most critical periods in human history and to feel all of the weight of it from the inside.

Apr 23 View on X →