OpenAI has launched two open-weight AI reasoning models, gpt-oss-120b and gpt-oss-20b, available on Hugging Face. These models aim to compete with other open-source AI models and are designed to be more accessible, with the smaller model capable of running on a consumer laptop.

While OpenAI had previously favored a closed-source approach, CEO Sam Altman acknowledged the company was on the wrong side of history regarding open sourcing. The release aligns with growing pressure from Chinese AI labs and encouragement from entities like the Trump Administration to promote AI aligned with democratic values.

The models excel at powering AI agents and calling tools like web search or Python code execution. They were trained using high-compute reinforcement learning, similar to OpenAI's proprietary models. However, they are text-only and cannot process images or audio.

OpenAI is releasing gpt-oss under the Apache 2.0 license, allowing enterprises to monetize the models without permission. While the models are state-of-the-art among open models, they hallucinate more than OpenAI's latest AI reasoning models, o3 and o4-mini. OpenAI delayed the release of its open models multiple times to address safety concerns, finding that gpt-oss may marginally increase biological capabilities but doesn't reach a high capability threshold for danger.

Cloudflare accused Perplexity of stealthily scraping websites, leading to a debate about AI agent access. The core question is whether an AI agent accessing a website on behalf of a user should be treated like a bot or a human.

Cloudflare tested Perplexity by blocking its AI crawler on a new website and then asking Perplexity about the site's content, which Perplexity answered. Cloudflare CEO criticized Perplexity, comparing its behavior to North Korean hackers.

Defenders of Perplexity argue that accessing sites on behalf of users is acceptable, questioning why an LLM accessing a site for a user should be treated differently than a web browser. Perplexity defended itself, claiming the behavior was from a third-party service.

The debate highlights the broader issue of bot activity reshaping the internet, with bots now outstripping human activity. LLMs are impacting website traffic, raising questions about whether websites should block AI agents that could potentially drive business.

WhatsApp is rolling out new features to combat scams, including safety overviews for group chats and alerts for individual chats with unknown contacts. The company has also taken down over 6.8 million accounts linked to scam centers. The new group chat feature provides key information and safety tips when someone not in your contacts adds you to a group. For individual chats, WhatsApp is testing alerts to caution users when starting a conversation with someone not in their contacts. WhatsApp also collaborated with OpenAI to disrupt scam efforts originating from a scam center in Cambodia, where scammers used ChatGPT to generate initial messages.

WhatsApp advises users to verify the legitimacy of requests, question if they make sense, and confirm the identity of individuals claiming to be friends or family.

The European Union's Artificial Intelligence Act (EU AI Act) is designed to establish a uniform legal framework for AI across EU countries, promoting the free movement of AI-based goods and services. It applies to both local and foreign companies involved in AI development and deployment.

The EU AI Act aims to foster trustworthy AI while protecting health, safety, fundamental rights, and environmental protection. It adopts a risk-based approach, banning unacceptable risk use cases, regulating high-risk uses, and applying lighter obligations to limited-risk scenarios.

The rollout of the EU AI Act began on August 1, 2024, with staggered compliance deadlines. As of August 2, 2025, it applies to general-purpose AI models with systemic risk (GPAI). Penalties for non-compliance can be significant, reaching up to €35 million or 7% of annual turnover for prohibited AI applications.

Some tech companies have expressed concerns about the EU AI Act, fearing it could hinder AI development and deployment in Europe. However, many companies, including Google, Amazon, and Microsoft, have signed the voluntary GPAI code of practice.

Google's NotebookLM, an AI note-taking app, is now available to Google Workspace for Education users of any age and consumers ages 13 and up. The update removes prior age restrictions, expanding access to AI-powered research tools to younger students.

Students can now use features like Audio Overviews, interactive Mind Maps, and Video Overviews to better understand class materials. Google says NotebookLM enforces stricter content policies for users under 18 and will not use their data for AI training.

The change follows OpenAI’s introduction of a study mode for ChatGPT, indicating increased competition in the AI education sector.

Cisco has disclosed a data breach that occurred on July 24th, resulting from a voice phishing (vishing) attack. A cybercriminal impersonated a trusted entity and tricked a Cisco representative into providing access to a third-party cloud-based customer relationship management (CRM) system.

The attacker then exported a subset of basic profile information belonging to Cisco.com users. The compromised data includes names, organization names, addresses, Cisco-assigned user IDs, email addresses, phone numbers, and account-related metadata such as account creation dates.

Cisco has not specified the number of affected users. This breach resembles a series of attacks targeting companies' Salesforce data, potentially impacting Cisco, a known Salesforce customer.