AI News — Sunday, June 7, 2026

9
OpenAI Unveils Lockdown Mode to Protect Sensitive Data from Prompt Injection Attacks

OpenAI has introduced 'Lockdown Mode' to enhance data security for enterprise users, specifically designed to protect sensitive information from prompt injection vulnerabilities.

TechCrunchproduct
9
Anthropic Launches Cowork: A Claude Desktop Agent for File Interaction Without Coding

Anthropic has released 'Cowork,' a new Claude desktop agent that allows users to interact with their files and perform tasks directly, eliminating the need for coding.

VentureBeatproduct
8
WWDC 2026 Expectations: Siri Revamp and Apple Intelligence Updates

Anticipation builds for WWDC 2026, with major announcements expected regarding a significant revamp of Siri and further updates to Apple's 'Apple Intelligence' initiatives.

TechCrunchindustry
8
Salesforce Rolls Out New Slackbot AI Agent Amidst Workplace AI Competition

Salesforce has launched an updated Slackbot AI agent, intensifying its competition with tech giants like Microsoft and Google in the rapidly evolving workplace AI market.

VentureBeatproduct
8
Dreaming: Better Memory for a More Helpful ChatGPT

OpenAI is enhancing ChatGPT's capabilities with a new 'Dreaming' feature, designed to improve its memory and make the AI more helpful and contextually aware over longer interactions.

OpenAI Blogproduct
7
A Blueprint for Democratic Governance of Frontier AI

OpenAI has published a comprehensive blueprint outlining strategies for the democratic governance of advanced 'frontier AI' systems, emphasizing safety and societal benefit.

OpenAI Blogindustry
7
Sriram Krishnan Departs Role as White House AI Advisor

Sriram Krishnan has announced his departure from his position as a key AI advisor to the White House, marking a notable change in the administration's AI leadership.

TechCrunchindustry
7
RobotValues: Evaluating Household Robots When Human Values Conflict

New research introduces RobotValues, a framework for evaluating household robots, specifically addressing scenarios where human values may conflict, to ensure ethical and safe deployment.

Hugging Faceresearch
7
Reinforcement Learning Elicits Contextual Learning of Unseen Language Translation

A new study demonstrates that reinforcement learning can enable AI models to achieve contextual understanding and translation of previously unseen languages, pushing the boundaries of multilingual AI.

Hugging Faceresearch
7
Language Models Transmit Behavioural Traits Through Hidden Signals in Data

Research published in Nature reveals that large language models can inadvertently transmit behavioral traits embedded as hidden signals within their training data, raising concerns about bias and influence.

Lobste.rsresearch
6
Rethinking Continual Experience Internalization for Self-Evolving LLM Agents

New research explores methods for LLM agents to continually internalize experiences, enabling them to self-evolve and improve their performance over time.

Hugging Faceresearch
6
Personal AI Agent for Camera Roll VQA

A new paper proposes a personal AI agent capable of performing Visual Question Answering (VQA) directly on a user's camera roll, offering intuitive image interaction.

Hugging Faceresearch
6
We Built a Coding Harness That Beats Frontier Models Using Open Ones (Open Beta)

A new coding harness has been developed that reportedly outperforms frontier AI models by leveraging open-source alternatives, now available in open beta.

Dev.toopen-source
6
AI Worm: A New Threat Vector for AI Systems

A new paper discusses the concept and potential implications of an 'AI Worm,' highlighting a novel threat vector for autonomous AI systems and their security.

Lobste.rsresearch
5
The Road Ahead in Autonomous Driving: The KITScenes Multimodal Dataset

Researchers introduce KITScenes, a new multimodal dataset designed to advance autonomous driving research by providing rich, diverse data for complex real-world scenarios.

Hugging Faceresearch