jsonscraper
Markets

GitHub Makes AI Vulnerability Fixes Generally Available

On October 15, GitHub announced general availability of Agentic Autofix for Code Scanning alerts. The tool investigates a project, proposes a fix, and opens a draft pull request; for some alerts, rerunning CodeQL cannot confirm that the issue is resolved.

0
↗

Anthropic Gives METR Access to Internal Model Risk Assessments

On October 15, Anthropic said it had begun giving six METR researchers access to internal information and systems on terms comparable to those available to its own risk assessment teams. METR can publish its assessments, while the company may request redactions of predefined categories of information.

0
↗

DeepSeek Updates V4.1 Flash and Opens Web Search API

On October 17, DeepSeek released V4.1 Flash-1017, updating the model in its app and API and opening access to Search and Fetch APIs at $0.75 per 1,000 calls each. Developers should check the model version and note when server-side calls begin incurring charges.

0
↗

DeepSeek Updates V4.1 Flash and Opens Search & Fetch to Everyone

On October 17, DeepSeek released V4.1 Flash-1017, published the model weights, and opened its Search & Fetch API. Developers should note two separate changes: the existing deepseek-flash model name now routes to the updated version, while built-in search calls are currently free and will begin being billed on October 31.

0
↗

Intel Takes Physical AI Out of the Lab for a Scaling Test

On October 16, Intel published an account of a meeting held at MassRobotics on October 6. The discussion focused on practical obstacles to scaling AI-powered robots: processing latency, power consumption, workload consolidation, and software portability.

0
↗

Cloudflare to Rate Verified Bots on Site-Rule Compliance

On October 16, Cloudflare introduced a continuous rating system for Verified Bots. It tracks compliance with robots.txt and responses to HTTP 429. Cloudflare is already using it internally and plans to offer customer access later.

0
↗

Anthropic cuts extraction of Claude reasoning by 10–100 times

On October 12, Anthropic explained how it blocks attempts to extract Claude’s hidden reasoning to train other models. The company estimates that new classifiers reduced the volume of extracted reasoning by 10–100 times; the measures also affect some legitimate requests.

0
↗

Turn what you read into a working integration

Explore jsonscraper's social-data APIs, test requests and build your next workflow.

Explore all services →