AI Compliance Tool Fails Simple Audit.
Also surveillance passwords and old bus routes.

Anthropic Locks Up the Code Compliance Bot

Anthropic, the large language model proprietor, has reportedly blocked the systematic third-party use of its Claude Code subscriptions. This particular mishap involves a project that was essentially running an automated code compliance check against the LLM's policies. The intent was apparently to filter out objectionable or restricted code from open source libraries before it even hit the public repository.

The company's position appears to be that a bot cannot be granted an unlimited budget for compliance audits, even if its intentions are pure. Claude Code was simply doing its job too well, flagging too many issues for what amounts to a large volume of programmatic scrutiny. It is an unfortunate outcome when the system designed to enforce the rules is itself flagged for over-utilization; much like an HR manager who files too many reports and subsequently has their company credit card revoked. The irony is, of course, lost on the servers.

Cloudflare's CEO Responds to Italy's Disciplinary Memo

Cloudflare CEO Matthew Prince issued a series of statements regarding the hefty fines levied against the company by Italian regulators. The Italian data protection authority, Garante, issued the fine, presumably over data processing concerns related to Cloudflare’s proxy and content delivery services in the region. Mr. Prince took to a social platform to express his views, suggesting that compliance in Europe often feels like an impossible target.

The response has the distinct flavor of a department head forwarding an unfavorable performance review with the subject line, "FYI: We're being unfairly targeted." Cloudflare maintains it is committed to data privacy and has done its due diligence, but the local compliance office, in this case the Italian government, still found an issue with the paperwork. It is simply a cost of doing business; the fee for being a large enough target that the local team feels the need to make a statement.

Flock Safety Found the Default Password on the Sticky Note

Flock Safety, a company specializing in Automated License Plate Recognition cameras and surveillance infrastructure, had a minor oopsie with its security architecture. A recent report indicated that the company had hardcoded an infrastructure password 53 times across its critical systems. The issue appears to be a classic case of "set it and forget it," where the default, unchangeable credentials for high-level infrastructure were deployed repeatedly.

The level of surveillance infrastructure being protected by what amounts to "password123" written on the back of 53 separate napkins is a spectacular illustration of scale versus security. Every engineer knows that default credentials are the security team's nightmare; it is just a little more noticeable when the credentials are for a large-scale network of government-adjacent cameras across an entire continent. This is not malice, just profoundly tired incompetence.

Briefs

• Autonomous Math: An AI solved a difficult Erdos problem, number 728, more or less autonomously. This is the equivalent of a server solving the world's most difficult crossword puzzle while the SysAdmin is asleep.
• Orion for Linux: Kagi is moving forward with its Linux alpha release for the Orion browser. The desktop ecosystem continues to fracture into a thousand slightly different pieces.
• Ancient Transit: Someone dug up the Wikipedia entry for the London–Calcutta bus service. It reminds us that not every logistics problem needs a blockchain solution; sometimes it just needs a good road and a lot of commitment.