AI News

Follow the latest AI and automation news

22 Apr 20265 Dhuʻl-Qiʻdah 1447 AH

Is Claude Code going to cost $100/month? Probably not - it's all very confusing

Anthropic has made an unannounced update to the Claude pricing page, altering the features of the Pro plan. The Claude Code feature, initially set to be included in the Pro plan, is now exclusive to Max plans starting at $100 per month. This update has sparked significant debate on platforms like Reddit and Twitter, with users criticizing the abrupt change. Amol Afsari, Anthropic's Head of Growth, noted that this shift is part of a small test for a limited group of new users.

Models

22 Apr 20265 Dhuʻl-Qiʻdah 1447 AH

Introducing OpenAI Privacy Filter

OpenAI has launched the Privacy Filter model, an open-weight solution designed to detect and redact personally identifiable information (PII) with high accuracy. This model is a significant tool for organizations handling sensitive data, ensuring compliance with privacy regulations. The Privacy Filter employs advanced natural language processing techniques, enabling precise identification of PII in text. This development is crucial for enhancing data privacy in various work environments, reducing the risks associated with data breaches.

Agents

21 Apr 20264 Dhuʻl-Qiʻdah 1447 AH

Humanoid data

Many individuals have been invited to participate in applications that pay them cryptocurrency for recording themselves performing simple tasks, such as placing food in a container. Robotics companies aim to utilize motion data to enhance humanoid robot capabilities, as they are deemed more suitable for human environments. With increasing investments in this sector, collecting real-world motion data has become essential for more effective robot training. Historically, companies relied on simulations to teach robots, but this method proved insufficiently accurate.

💡

World models

How could world models enhance AI performance in everyday tasks?

Agents

Agent orchestration

21 Apr 20264 Dhuʻl-Qiʻdah 1447 AH

Agents

21 Apr 20264 Dhuʻl-Qiʻdah 1447 AH

Agent orchestration

The development of multi-agent AI tools, such as OpenClaw, has accelerated, attracting the interest of major companies like Nvidia and Tencent. These tools enable teams of agents to collaborate, facilitating the completion of more complex tasks. For instance, Anthropic's Claude Code allows the coordination of multiple programming agents simultaneously, enhancing productivity. These advancements are transforming developers into project managers, enabling them to delegate and monitor tasks more effectively.

Business

China’s open-source bet

21 Apr 20264 Dhuʻl-Qiʻdah 1447 AH

Business

21 Apr 20264 Dhuʻl-Qiʻdah 1447 AH

China’s open-source bet

Silicon Valley AI companies typically keep their APIs proprietary, charging for access. In contrast, leading Chinese AI labs are releasing models as open-weight packages, allowing developers to adapt and run them independently. This shift gained traction after DeepSeek's R1 model launched in January 2025, matching top U.S. systems at lower costs, narrowing the competitive gap.

Governance

Weaponized deepfakes

21 Apr 20264 Dhuʻl-Qiʻdah 1447 AH

Governance

21 Apr 20264 Dhuʻl-Qiʻdah 1447 AH

Weaponized deepfakes

Experts have warned for years that deepfake technology, which creates fake videos and images, could be used harmfully. With the increasing accessibility of generative models, anyone can easily distort reality in ways that are hard to detect. Studies indicate that 98% of deepfakes are pornographic in nature, with 99% involving women. The risks include the use of deepfakes in political propaganda, where fake images and videos are employed to influence public opinion. For instance, the Trump administration shared AI-generated content, highlighting how this technology can be used to distort facts.

Governance

Supercharged scams

21 Apr 20264 Dhuʻl-Qiʻdah 1447 AH

Governance

21 Apr 20264 Dhuʻl-Qiʻdah 1447 AH

Supercharged scams

The launch of ChatGPT in late 2022 has led to an increase in the use of large language models by criminals, who are employing these tools to craft malicious emails, including spam and targeted attacks. This technology has accelerated their operations, enabling the creation of fake videos and the analysis of stolen data. Criminal gangs are utilizing AI tools to identify vulnerabilities in networks and generate ransom notes, facilitating the targeting of a larger number of victims. Interpol has warned that fraud centers in Southeast Asia are adopting inexpensive AI tools to quickly target new victims.

Models

OpenAI Beefs Up ChatGPT’s Image Generation Model

21 Apr 20264 Dhuʻl-Qiʻdah 1447 AH

Models

21 Apr 20264 Dhuʻl-Qiʻdah 1447 AH

OpenAI Beefs Up ChatGPT’s Image Generation Model

ChatGPT Images 2.0 has been launched, showcasing significant improvements in detailed image generation and text display. Tests indicate that the new model surpasses its predecessor in image quality, making it a powerful tool for users. However, it still struggles with non-English languages, limiting its applicability in diverse markets. These enhancements stem from advanced techniques in image and text processing, with a focus on improving detail accuracy. The new model reflects progress in artificial intelligence technologies, making it an attractive option for developers and designers.

Agents

How to evaluate the performance of AI agents?

21 Apr 20264 Dhuʻl-Qiʻdah 1447 AH

Agents

21 Apr 20264 Dhuʻl-Qiʻdah 1447 AH

How to evaluate the performance of AI agents?

AI agencies face multiple evaluation challenges, as traditional testing methods are insufficient. These agencies are non-deterministic, meaning identical inputs can yield different outputs, complicating performance assessment. When developing AI agencies, three main challenges must be addressed: evaluating pathways rather than just outputs, the difficulty of defining successful performance, and the need for continuous evaluation. These factors necessitate the use of varied assessment methods for each quality dimension.