AI News Today: Claude Code Leak Beats Turbine Deal

AI news today, 9 November 2026: the story that deserves your attention is the Claude Code "Yes, and don't ask again" leak, which Apiiro counted at 2,539 distinct secrets across 1,994 repositories in research circulating 7-9 November. It beats Anthropic's gas turbine purchase because it demands action this week.

TL;DR

• Choosing "Yes, and don't ask again" in Claude Code saves the full approved shell command, secrets included, as an allow rule in a local settings file (Apiiro).
• The default protection sits in a global git excludes file, so the project's own .gitignore never covers it (Apiiro).
• Anthropic took over orders for 1.7 GW of Caterpillar generators worth $3.4B, about Rs 3.19 lakh crore at Rs 93.8 per dollar (approximate conversion) (CNBC).
• Ramp's index shows token-spend growth stalling; price cuts, not weak demand, appear to explain it (TechCrunch).
• Anthropic's IPO marketing could begin on 10 November, with valuation talk reportedly lower than earlier in the month (Investing.com).
• Last verified: 9 November 2026.

What is the main AI news today?

The main AI news today is the Claude Code secret leak, followed by the turbine order, the Ramp spending data, the IPO timing and an agent-memory research post. The table ranks them by the action they demand now.

The ranking is our editorial judgement; for prior coverage see DeepMind agents weakening 1024-bit RSA and the 5 November digest.

How did Claude Code leak 2,539 secrets?

Claude Code leaked secrets because its "don't ask again" option stores the entire approved command, including any password or key typed inside it, as a permanent allow rule. That rule is written to .claude/settings.local.json. Claude Code's default safeguard adds that path to the user's global git excludes file, but the repository's .gitignore does not carry it, so clones, CI runners and teammates get no protection (Apiiro).

Apiiro found 104,185 public copies across 103,617 repositories; 1,122 of them, about 1.1%, held a real secret. Adding git history, the shared .claude/settings.json and saved agent chat logs lifts the total to 2,539 (Apiiro).

Persistence makes it worse: the median age of the file in an affected repository was 217 days, and a GitHub report filed on 11 August 2025 was closed as "not planned" on 3 January 2026 (Apiiro).

Which secrets leaked, and why did scanners miss some?

Mostly database passwords and cloud credentials - and scanners missed nearly a fifth of the affected files. The breakdown of the 1,122 files included 379 database passwords, 122 Supabase service-role keys, 43 SSH logins, 42 Git host or npm tokens and 23 AWS keys (Apiiro).

Gitleaks flagged 564 of the 1,122 files and TruffleHog flagged 643, while 215 files, around 19%, were caught by neither. The misses were passwords typed after patterns such as PGPASSWORD=, sshpass -p or mysql -p, or placed in request headers (Apiiro).

Two caveats: Apiiro did not test whether keys were live, though one notified company confirmed its leaked key was (The Hacker News); and Check Point found 428 public npm packages shipping the file, 33 with credentials (Cloud Security Alliance note).

What should teams do about the Claude Code leak today?

Rotate first, then clean up, then change habits. Apiiro's priority order, reproduced in plain steps:

1. Rotate every credential the file has ever held, including values only present in git history.
2. Run git rm --cached .claude/settings.local.json and add the path to the repository's own .gitignore.
3. Delete any allow rule that contains a literal secret.
4. Reference secrets as $VAR values supplied by a secrets manager.
5. Approve patterns such as Bash(npm run:*) instead of literal commands.

Rotation comes first: deleting the file does not remove it from history (Apiiro). Our guides to message threads and cheaper subagent status updates cover other Claude Code settings worth reviewing.

Does the Claude Code leak matter for Indian development teams?

Yes. The global-exclude protection lives on one developer's laptop and does not travel with the repository, so teams that share workspaces between machines - outsourced teams and global capability centres especially - lose the only guard when a settings file moves or comes from a fresh clone. Apiiro's sample included an IT services firm whose medical-records application had its database owner password sitting in a public allowlist (Apiiro). Indian teams should add the .gitignore entry to their project templates and scan existing repositories this week.

Why did Anthropic buy $3.4 billion of gas turbines?

Anthropic bought the turbines because grid-connection queues and manufacturing lead times are the bottleneck for new data centres; taking over queued orders skips part of the wait. In late September it took over 1.7 GW of heavy-duty Caterpillar generator commitments from OnePWR Solutions of San Antonio for $3.4B, roughly Rs 3.19 lakh crore at Rs 93.8 per dollar (approximate conversion) (CNBC).

OnePWR ordered the equipment about two years earlier and could not finance it; Anthropic had planned to buy power, then took over the orders instead - engineering and construction included, no staff transfer (TechCrunch). The deal sits alongside $895.4B of disclosed purchase commitments in its S-1 (CNBC).

Is AI spending stalling, and when could Anthropic list?

Spending growth has slowed, but the evidence points to price cuts rather than falling usage: Ramp's AI Index showed token spend growing 1.5% across August and September, against 42% growth in July, then re-accelerating 8.9% between 23 September and 23 October (TechCrunch), while volumes kept rising. For Indian startups, that is a plausible reason API bills fell in September and October even as usage grew.

On timing, marketing could begin on Tuesday 10 November, with pricing likely the week of 16 November, before Thanksgiving on 26 November. Reported valuation talks fell from $1.5-2 trillion earlier in the month to $1.1-1.3 trillion after the Nasdaq 100 dropped 14% (Investing.com). Treat these as reported expectations, not confirmed terms.

What did Anthropic publish about agent memory?

Anthropic's engineering post "The components of agent memory" reports that a memory store lifted Opus 5.5's mean reward on a CL-Bench database task from 0.655 to 0.838, while Sonnet 4.6 moved only from 0.330 to 0.364; "dreaming" sessions across 128 runs on a 420M-token corpus raised Opus 5 agents' graded criteria from 13% to 48% (Anthropic). Dreams are a research preview, unsuitable for zero-data-retention or HIPAA deployments. Note the irony: persistent agent memory is exactly where secrets can accumulate, so apply the same hygiene.

FAQ

Q: Which AI story matters most on 9 November 2026?
A: The Claude Code leak - it needs action now, and Apiiro counted 2,539 distinct secrets in 1,994 repositories.

Q: Is my project affected if I never chose "don't ask again"?
A: Probably not through that route, but check anyway: Apiiro found 1,122 of 104,185 public copies held a real secret.

Q: Should I delete the file or rotate the keys first?
A: Rotate first - removing the file does not erase git history, and Apiiro's fix order puts rotation ahead of cleanup.

Q: How much did Anthropic pay for the turbines?
A: It took over 1.7 GW of orders for $3.4B, about Rs 3.19 lakh crore at Rs 93.8 per dollar (approximate conversion).

Q: Is enterprise AI demand falling?
A: Not clearly: Ramp measured 1.5% token-spend growth in August and September, but its economist points to price cuts, and volumes kept rising.

For related security payouts, see our note on the OpenAI bug bounty increase.

This article was prepared with AI assistance and checked against the primary sources linked above. Corrections are logged on this page as they arrive.