🧭 Code with Claude London Opens Today — Three Tracks, Live Keynote & European Developer Focus

The second stop of Anthropic's 2026 developer conference circuit opens in London today (19 May), running 7:30 AM–7:30 PM BST. Unlike the San Francisco launch event on May 6, the London edition is structured around three simultaneous stage tracks tailored to different practitioner audiences: Model Research (capabilities and future directions), Claude Platform (production-grade agent development), and Claude Code (scaling agents and tooling infrastructure). Day 1 keynote and breakout sessions are streamed live on Anthropic's site.

Keynote speakers and agenda

The morning keynote (9:00–10:00 AM BST) features Angela Jiang, Boris Cherny, Cat Wu, Katelyn Lesse, and Lisa Crofoot from Anthropic's product, research, and engineering leadership. The afternoon programme includes:

• "What's new in Claude Code" — latest tooling updates from the Claude Code team
• "Stop babysitting your agents" — patterns for running fully autonomous multi-agent pipelines
• "The prompting playbook" — structured prompting techniques from the Claude Platform team
• "The thinking lever" / "The capability curve" — Research track deep-dives on extended thinking and model progression
• "How to get to production faster with Claude Managed Agents" — hands-on workshop

European ecosystem on stage

Guest companies presenting AI-native development case studies include Spotify, Lovable, monday.com, and Doctolib — providing a distinctly European lens on enterprise and consumer AI deployment that complements the US-heavy SF programme.

🧭 Claude Security Enters Public Beta — Full-Codebase Vulnerability Scanning and Auto-Patching with Opus 4.7

Claude Security is now in public beta for Claude Enterprise customers (Team and Max access coming soon). Available directly from the Claude.ai sidebar or at claude.ai/security, it uses Opus 4.7 to scan entire codebases for vulnerabilities — reasoning about code the way a security researcher would, rather than matching patterns.

How it works

Claude Security traces data flows and reads interactions across files and modules to detect complex, context-dependent flaws that signature-based scanners miss. Each finding comes with:

• Confidence ratings to reduce false-positive noise
• Severity classification and reproduction steps
• Automated patch generation — fixes are applied directly through Claude Code, so remediation stays inside your existing workflow
• Scheduled and targeted scans — run against a full repo, a directory, or a single PR
• Audit trail — findings can be dismissed with documented reasons; results export to CSV or Markdown, with webhooks for Slack and Jira

Partner integrations

Beyond the native claude.ai/security interface, Anthropic has partnered with CrowdStrike, Palo Alto Networks, Microsoft Security, SentinelOne, TrendAI, and Wiz to embed Opus 4.7 scanning inside the security platforms enterprise teams already run. Services partners — Accenture, BCG, Deloitte, Infosys, and PwC — are also building Claude Security into their managed security offerings.

# From Claude.ai sidebar → Security → New Scan
# Or direct URL:
# https://claude.ai/security

# After findings appear, click "Fix with Claude Code"
# to open a patch branch in your connected repo

🧭 The Anthropic Institute Publishes Its Four-Pillar Research Agenda on AI's Societal Impact

The Anthropic Institute (TAI) — Anthropic's internal research arm focused on AI's real-world societal effects — published its formal research agenda on May 7. Unlike the safety and alignment teams that evaluate models before deployment, TAI operates at the population level: studying what happens after Claude-scale systems interact with millions of users, organisations, and economies simultaneously. The agenda is explicitly described as a "living document" that will evolve as findings land.

The four pillars

• Economic Diffusion — How powerful AI systems reshape jobs, productivity, and who captures the gains. TAI will publish higher-frequency, more granular data from the Anthropic Economic Index, tracking adoption patterns across countries, firms, and sectors in near real-time.
• Threats and Resilience — The dual-use reality: AI advances that accelerate beneficial research can equally accelerate cyber, bio, and surveillance threats. TAI researchers are building early-warning systems and modelling how societies can strengthen defensive infrastructure faster than threat surfaces expand.
• AI Systems in the Wild — What actually changes when a large population relies on shared AI models for information and reasoning. Research covers epistemology shifts, critical-thinking degradation, human–AI team dynamics, and governance mechanisms for increasingly autonomous agents.
• AI-Driven R&D — The recursive self-improvement question: how AI accelerates scientific research, and under what conditions AI systems could meaningfully contribute to developing successor AI systems. TAI is measuring research velocity and modelling intervention points for controlling acceleration rates.