AI attackers are getting harder to spot after a year of data

Anthropic studied 832 banned accounts and found AI is making less-skilled attackers more dangerous, and the old ways of assessing threat level don't work anymore.

Why it matters: The share of actors classified as medium-risk or higher jumped from 33% to 56% in a single year. Post-compromise techniques that used to require real skill can now be handed off to an AI agent, so the skill gap between amateur and advanced attackers is shrinking fast.

• The clearest signal of a high-risk actor is no longer the number of techniques they use (low vs. high skill actors are nearly identical: 16 vs. 20 techniques on average). It's whether they build scaffolding that lets the model chain attack stages together autonomously.
• MITRE ATT&CK has no ID for agentic orchestration, which is exactly the capability that defines the most dangerous actors. Anthropic is in talks with MITRE to update the framework.

Go deeper: https://www.anthropic.com/news/AI-enabled-cyber-threats-mitre-attack

Uber caps Claude Code and Cursor spending at $1,500 per engineer per month

After burning through its 2026 AI budget in four months, Uber set a per-tool monthly token limit — a pragmatic call that also reveals how much companies expect to get out of these tools.

Why it matters: The cap implies Uber is willing to spend up to $36,000 per engineer per year on coding agents, roughly 11% of the median US comp package for their engineers. That's a real number, not a pilot.

• Simon Willison notes the limit is tool-specific (Cursor and Claude Code counted separately), which suggests Uber is still all-in on agentic coding — just with guardrails.
• The policy quietly kills "tokenmaxxing" leaderboards that had some teams competing to burn as many tokens as possible.

Go deeper: https://simonwillison.net/2026/Jun/3/uber-caps-usage/#atom-blogmarks

Anthropic adds formal tiers to its partner network

Three months after launching the Claude Partner Network with a $100M commitment, Anthropic is adding a tiered structure and a public partner directory so enterprise buyers can tell who has actually shipped Claude in production.

Why it matters: More than 40,000 firms applied to join and 10,000 consultants earned certifications. The new Services Track — Select, Preferred, Global Premier — measures deployed customers and certified practitioners, not revenue sent to Anthropic, which is tracked separately.

• The six biggest consulting firms (Accenture, Cognizant, Deloitte, KPMG, Infosys, PwC) are in, with employee rollouts ranging from 30,000 to 470,000 people.
• Partners can now connect the Claude Partner Hub to Claude via MCP, letting a firm ask Claude where it stands against the next tier or check deal registration status directly.

Go deeper: https://www.anthropic.com/news/services-track-partner-hub