Retter Reboot #52

OpenAI Warns AI Browsers May Always Be Vulnerable to Prompt Injection Attacks

OpenAI has cautioned that AI-powered browsers and agents may never be fully immune to prompt injection attacks, due to their need to interpret and act on untrusted web content. The company explained that as AI systems browse, summarize, and execute tasks across the open internet, malicious instructions embedded in pages can manipulate model behavior in hard-to-detect ways. While mitigations and safeguards can reduce risk, OpenAI suggests the issue is structural rather than temporary, highlighting a long-term security challenge for autonomous AI agents.

Source: Techcrunch

Anthropic Releases Bloom, an Open-Source Agentic Framework for Evaluating Frontier AI Models

Anthropic has open-sourced Bloom, a new agentic framework designed to automate behavioral evaluations of frontier AI models. Bloom enables researchers and developers to run large-scale, repeatable tests that assess how models behave across complex, multi-step scenarios, including safety, reliability, and decision-making under pressure. By making Bloom open source, Anthropic aims to standardize and accelerate rigorous AI evaluation practices as models become more autonomous and capable.

Source: Marktech Post

Alphabet to Acquire Intersect for $4.75 Billion in Cash

Alphabet has agreed to acquire Intersect in an all-cash deal valued at $4.75 billion, expanding its footprint in AI infrastructure and enterprise technology. The acquisition is expected to bolster Alphabet’s capabilities in data management and large-scale systems that support advanced AI workloads, as competition intensifies among big tech firms to secure the foundational layers of AI development. The move underscores Alphabet’s continued strategy of using targeted acquisitions to strengthen its AI and cloud ecosystem.

Source: WSJ  

Cursor Continues Its Acquisition Spree with Graphite Deal

Cursor has acquired Graphite as part of its ongoing acquisition push to expand its AI-powered developer tooling ecosystem. The deal brings Graphite’s code review and collaboration capabilities into Cursor’s AI-first coding environment, strengthening its position as an end-to-end platform for modern software development. The acquisition highlights intensifying competition among AI coding toolmakers as they race to bundle editing, review, and collaboration into unified workflows.

Source: Techcrunch

Z.ai Releases GLM-4.7, Claiming GPT-5.1-Level Performance with Agentic Reasoning

Chinese AI lab Z.ai has unveiled GLM-4.7, a new large language model it claims reaches performance parity with GPT-5.1 while preserving internal “thinking” capabilities optimized for agent-based workflows. The model is designed to support long-horizon reasoning, tool use, and autonomous task execution without sacrificing response quality. GLM-4.7 positions Z.ai as a serious contender in the global race to build frontier models tailored for agentic and enterprise-grade AI systems.

Source: Winbuzzer

OpenAI Updates ChatGPT to Sound Warmer and More Enthusiastic

OpenAI has rolled out an update to ChatGPT focused on adjusting its conversational characteristics, making responses sound warmer, more enthusiastic, and more engaging. The change is part of ongoing tuning to improve how the assistant communicates tone, encouragement, and emotional nuance without altering core capabilities. OpenAI says the update reflects user feedback and aims to make interactions feel more natural as ChatGPT is used more frequently in everyday contexts.

Source: The Verge

Meta Developing New AI Image and Video Model Code-Named “Mango”

Meta is reportedly developing a new AI model for image and video generation internally code-named “Mango,” as it ramps up competition with OpenAI, Google, and Runway in multimodal AI. The model is expected to improve visual quality, motion consistency, and creative control, supporting both still images and video generation across Meta’s platforms. The effort reflects Meta’s broader strategy to build in-house foundation models that power creative tools for Instagram, Facebook, and future immersive experiences.

Source: WSJ 

Amazon’s Alexa Expands AI Capabilities Through Partnerships with Angi, Expedia, Square, and Yelp

Amazon has expanded Alexa’s AI assistant capabilities by integrating it with third-party services including Angi, Expedia, Square, and Yelp. The new partnerships allow Alexa to help users book services, make reservations, manage payments, and discover local businesses directly through conversational interactions. This move positions Alexa as a more action-oriented assistant, shifting from simple voice queries toward completing real-world tasks across commerce, travel, and local services.
Source: Techcrunch 

Google Gemini Is Getting an AI Video Detector

Google is adding an AI-powered video detection feature to Gemini, designed to help identify AI-generated or manipulated video content. The tool will analyze visual patterns, motion artifacts, and metadata signals to flag synthetic or altered footage, addressing growing concerns around deepfakes and misinformation. The move reflects Google’s broader effort to build trust and transparency into its AI ecosystem as generative video tools become more widespread.
Source: How To Geek

Instacart Ends AI-Based Pricing Tests After Shutting Down Eversight

Instacart has ended its experiments with AI-driven dynamic pricing following the shutdown of Eversight, the pricing technology firm it acquired in 2022. The company said it will no longer test algorithmic price optimization tools that adjusted prices in real time, amid growing regulatory scrutiny and consumer sensitivity around price transparency. The decision signals a pullback from AI-led pricing experimentation in grocery and retail commerce.

Source: The Verge

Italy Orders Meta to Halt WhatsApp Terms That Restrict Rival AI Chatbots

Italy’s competition watchdog has ordered Meta to stop enforcing WhatsApp terms that limit or bar the use of rival AI chatbots, citing concerns over anti-competitive behavior. Regulators argue that the restrictions could unfairly advantage Meta’s own AI services and hinder competition in the rapidly evolving AI assistant market. The decision adds to growing regulatory pressure in Europe on how big tech companies integrate AI features into dominant platforms.

Source:Reuters