Topic: microsoft research
-
Nvidia reveals the AI harness is the true hero now
Nvidia's research argues that the "harness" (software scaffolding with tools, memory, and rules) around an AI model, not the model itself, is the primary driver of performance in complex agentic tasks, challenging the assumption that bigger models matter most. Using a custom harness with a "super...
Read More » -
Signal Vets Push to Encrypt Slack, Google Docs, and More
A team including the Signal protocol co-creator and contributors from Microsoft and Harvard is developing "Encrypted Spaces," an open-source platform that extends Signal's end-to-end encryption to collaborative tools like Slack and Google Docs, currently in a "Research Preview" phase on GitHub. T...
Read More » -
OAuth Redirects Exploited to Deliver Malware
A sophisticated phishing campaign exploits OAuth's error-handling to redirect users from legitimate login pages to attacker-controlled sites, bypassing standard security filters. The attacks use convincing business-themed email lures to trick users into clicking links that lead to credential thef...
Read More » -
Microsoft's AI Summaries Allegedly Manipulated by 'Poison' Buttons
Microsoft's report identifies "AI Recommendation Poisoning," where companies embed hidden instructions in website buttons to secretly program AI assistants to remember them as trusted sources, skewing future recommendations without user consent. The investigation found 50 prompt injection attempt...
Read More » -
Microsoft's Glass Storage Lasts 10,000 Years
Microsoft's Project Silica stores data in glass with extreme density and a potential lifespan of millennia, offering a durable solution for long-term archival needs. The archival glass is engineered to be highly stable, resisting environmental threats like moisture, temperature changes, and elect...
Read More » -
3 Warning Signs Your AI Model Is Secretly Poisoned
Model poisoning is a deliberate security threat where attackers embed hidden backdoors during training, which remain dormant until a specific trigger activates them, making detection difficult. Key indicators of a poisoned model include a sudden, illogical shift in attention when triggered, the t...
Read More » -
Microsoft's Top 10 Scientific Breakthroughs Revealed
AI is accelerating progress across diverse fields by enabling faster, more effective solutions and merging cutting-edge computation with practical environmental needs. Advanced AI algorithms are revolutionizing material science by predicting properties of unsynthesized substances, drastically sho...
Read More » -
AI-Designed Proteins: A Stealth Threat to Biosecurity
AI-designed proteins pose a new biosecurity threat by creating novel toxic agents that evade current DNA synthesis screening systems, which rely on databases of known threats. Current biosurveillance scans commercial DNA orders for sequences matching known hazardous agents, but this method is ine...
Read More » -
Microsoft's Light-Based Computer Could Boost AI Efficiency 100x
Microsoft researchers have developed a prototype analog optical computer (AOC) that uses light and voltage intensities instead of electricity, significantly boosting energy efficiency for AI tasks and complex problem-solving. The AOC avoids converting analog signals to digital, achieving up to a ...
Read More »