Schools must opt into Google AI products, and search counts as part of that AI portfolio under the new K-12 policy announced this week.

Logged-in student accounts get search only when schools turn it on, then choose between standard AI features and the Guided Learning version.

Schools can disable Gemini apps while leaving search active, so the AI-free and AI-lite options both exist without touching Gmail or Docs.

Google will not build a toggle to turn off AI tools in search, even though competing search engines already offer exactly that kind of setting.

Google lost the DOJ search antitrust case, which gives rivals only limited access to its web index, and that constrains who can offer an alternative.

Hy4 scored 3.04 on a blind financial evaluation with 213 professionals, edging out DeepSeek-V4.1-Flash-1017 at 2.97 and Kimi K3.1 Pro at 2.68.

The model's hallucination rate fell from 12.95% in Hy4 Preview to 7.2%, while errors from unsupported assumptions dropped from 23.1% to 5.6%.

On Tencent's internal Investment Research Agent Benchmark, Hy4 improved 12.03 percentage points over its Hy4 Preview predecessor, a big jump for a single release.

Game development gains came from work with Tencent Interactive Entertainment Group developers, with GameDev up 19.1% and GameLogic up 9.8% against prior benchmarks.

Tencent open-sourced HyCell for single-cell representations and HyPDE for physics fields, both built on Hy4 and Hyra 2.0 foundations.

Claude sees up to two years of transaction history, along with balances, credit limits, bank name, account names, account type, and the last four digits of each account number.

The Personal Finance beta is limited to US subscribers on Claude Pro, Max, and Ultra plans, rolling out now as a preview rather than a general release.

Access runs through Plaid, and Claude never needs your banking password, which removes one of the biggest credential-sharing risks in consumer fintech.

Claude cannot move, manage, or invest money, so the worst case is bad advice rather than unauthorized transfers, which matters for compliance teams.

OpenAI introduced a similar ChatGPT finance feature in May, making this the second major AI assistant to request direct access to bank data.

Midterm results were called a total repudiation of the AI field, with Democrats taking control of both the House and Senate chambers in Congress.

Seventy-six percent of voters polled on Tuesday said they opposed data centers being built in their communities, a figure that cuts across party lines.

43 out of 44 local data center moratoria on Tuesday's ballots passed, showing residents will vote directly to block construction projects near them.

Governor-elect Gina Hinojosa plans to issue a moratorium on all data center approvals until the state legislature passes laws governing their construction and operation.

Yorkville Ives & Co. analysts advise tech investors to prepare for major hurdles for AI legislation and a measurable slowdown in data center construction next year.

Rivet Actors have a p99 cold start of 4.99 ms, compared with 1.70 s for sandboxes, a difference of roughly 340 times for workloads that spin up frequently.

Each actor uses 72.4 KB of memory, while a sandbox needs 512 MiB, so a single server can hold hundreds of thousands of actors at once.

Storage for Rivet Actors costs 10.4 times less than replicated SSD storage, which matters once you're keeping state for thousands of idle agents.

Rivet works with Amazon S3, Google Cloud Storage, Azure Blob Storage, and any S3-compatible store, and the object storage bucket is its only external dependency.

Rivet lifts the 128 MB memory cap and storage limits of Cloudflare Durable Objects, and it runs on vanilla Node.js or Bun rather than a custom V8 runtime.

GPT-6 Cyber scored 48.5 percent on OpenAI's ExploitGym evaluation, compared with 32.1 percent for GPT-5.6 Cyber on the same benchmark tests.

The model stayed within authorized scope over 99 percent of the time, with only 0.07 percent of cases going beyond the intended boundaries of the training exercise.

GPT-6 Cyber found 81 previously unknown high-severity vulnerabilities in Firefox, and helped researchers build working exploits for 948 Linux kernel vulnerabilities, including 588 new bugs.

OpenAI classified GPT-6 Cyber as "Critical" under its Preparedness Framework, and it is initially available only through the API and Codex to select Daybreak Red customers.

Critical findings on core products now earn up to $30,000, or up to $51,577 with bonuses, while exceptional findings still qualify for up to $100,000 or $172,000 with bonuses.