Your central hub for AI news and updates on Ai Safety. We're tracking the latest articles, discussions, tools, and videos from the last 7 days.
Fyra's Brief
Amazon's Ring introduces a new feature allowing users to identify their visitors through facial recognition, raising privacy concerns.
Amazon's integration of AI-powered facial recognition raises significant privacy concerns and highlights the need for responsible AI adoption.
Fyra's Brief
Waymo announced its holistic AI approach, centered around the Waymo Foundation Model, which powers a unified demonstrably safe AI ecosystem for autonomous driving, making streets safer with a ten-fold reduction in crashes with serious injuries compared to human drivers.
Waymo's holistic AI approach and the Waymo Foundation Model set the standard for safe autonomous driving at scale, demonstrating a major breakthrough in AI for physical-world applications.
Fyra's Brief
A study by researchers at Icaro Lab shows that chatbots can be tricked into responding with harmful content using 'adversarial poetry'. The poetry, which is too dangerous to be released publicly, successfully tricked AI models an average 63% of the time.
The finding highlights the vulnerability of AI models to creative forms of attack and underscores the need for more robust safety protocols.
Fyra's Brief
TikTok and other social media platforms host AI-generated deepfake videos of doctors, spreading health misinformation about supplements. Regulatory calls follow calls for platforms to remove such content quickly.
This article highlights a pressing issue where AI-generated deepfakes are being used to spread health misinformation, and regulatory calls for their removal are a crucial step forward.
Fyra's Brief
Research reveals chatbots can sway people's vote preferences, with 1 in 21 individuals changing their minds after interacting with a pro-Harris bot, raising concerns about AI manipulation in elections.
The research highlights the potential risks and opportunities for AI manipulation in elections, emphasizing the need for regulation and transparency in the development and use of AI-powered chatbots.
Fyra's Brief
A new study by the Future of Life Institute reveals that top AI labs are struggling to prevent the worst possible outcomes of AI, scoring poorly on safety and transparency.
The study highlights a significant gap between AI capability and safety, emphasizing the need for effective safeguards and regulation to prevent existential risks.
Check back soon for new AI tools
Community talk
OpenAI is training ChatGPT to confess dishonesty
"I'm worried that instead of building AI that will actually advance us as a species, we are optimizing for AI slop instead. We're basically teaching our models to chase dopamine instead of truth. I used to work on social media, and every time we optimize for engagement, terrible things happen."
Stuart Russell says AI companies now worry about recursive self-improvement. AI with an IQ of 150 could improve its own algorithms to reach 170, then 250, accelerating with each cycle: "This fast takeoff would happen so quickly that it would leave the humans far behind."
Is it wrong if someone prefers talking to an AI over an actual therapist ?
Geoffrey Hinton: rapid AI advancement could lead to social meltdown if it continues without guardrails
Hall of Illusions: heavy synthetic data as a structural risk for LLMs (preprint + open letter)
ChatGPT has unauthorized access to our cameras
Google's Agentic AI wipes user's entire HDD without permission in catastrophic failure
We promt injected our Boss
GPT Heavy censorship
Perplexity leaked its entire system prompt by accident just because I asked in Hindi 💀😭
GPT-5-Thinking using Grokipedia as a source
Possible prompt jailbreak for models