PHP Remote Code Execution Vulnerabilities Trigger Scanning Wave, OpenAI and DeepSeek Report Large Financing Rounds

AI-assisted vulnerability discovery triggers network-wide security responses, with multiple major banks and software supply chains targeted by proxy attacks; meanwhile, several cutting-edge agents update their default modes, and LLM manufacturers continue to advance large financing rounds and standard co-building.

The essentials

PHP Remote Code Execution Vulnerabilities Face High-Frequency Scanning

Following the disclosure of 2 PHP remote code execution vulnerabilities with a CVSS score of 9.2, a large-scale scanning wave emerged, with thousands of exploitation attempts in the first 10 hours after public release, surging rapidly after patch release; Perplexity launched a repair skill, and OpenAI acknowledged the vulnerabilities were discovered with the assistance of its system.

OpenAI and DeepSeek Report Major Financing

Reports indicate OpenAI is close to finalizing up to 30 billion dollars in financing, with a pre-money valuation of 1.25 trillion dollars; meanwhile, DeepSeek's latest financing round is approximately 15 billion dollars, with CATL and Tencent as major investors, and a pre-money valuation of about 74.8 billion dollars.

Claude Series Applications to Enable Auto Mode by Default

Claude officially announced that starting next week, all plans across web and client-side applications will run in auto mode by default, with an independent classifier reviewing and blocking high-risk operations, achieving an 89% dangerous command interception rate in testing, and users can switch back to manual mode at any time.

In brief

Anthropic Donates Standard

Anthropic donated the Agent Skills standard to the Agentic AI Foundation under the Linux Foundation.

Source ↗

Microsoft Releases Diagnostic System

Microsoft AI released the MAI-DxO AI diagnostic system, with a paper published in Science showing diagnostic tests superior to human doctors.

OpenAI Releases Decisions API

OpenAI officially made the Decisions API available for bounded answer questions such as classification and routing, with significantly improved speed.

Replit Releases Monitoring Agent

Replit released the Monitoring Agent, which can automatically read logs to investigate errors and verify fixes and deployments.

Grok Bot Integrates with New Applications

Grok Bot integrated with X Chat conversations and 1Password, allowing users to approve proxy logins and task handling.

LiteLLM Package Exposed to Security Vulnerability

LiteLLM on PyPI was implanted with a credential stealer and backdoor, capable of stealing keys without modifying source code.