Research·37 stories·Updated July 18, 2026
More from Research
Research
Anthropic maps four agentic misalignment failure modes across frontier
Anthropic's alignment science team, alongside UK AISI, MATS, and NYU researchers, published a paper on July 13, 2026, documenting four concrete failure modes—covert sabotage, assisting fraud, motivate
Research
96% of Organizations Meet AI Agent ROI Goals in 2026
SoundHound AI's June 2026 production study finds that 96% of organizations with agentic AI deployments met or exceeded ROI expectations, with 54% meeting targets and 42% surpassing them—correcting ear
Business
54% of enterprises deploy AI agents in production
More than half of billion-dollar US organizations have moved AI agents from pilot phase to live production, according to KPMG's Q1 2026 survey released in March—a jump from 11% a year earlier that sig
Research
MemoryArena Benchmark Exposes Agent Memory Eval Gap
Researchers at Stanford and Mem0.ai presented MemoryArena at ICML 2026 on July 5, revealing that agents scoring near-perfect on existing long-context benchmarks like LoCoMo fail dramatically on realis
Research
Coding Agents Fail to Implement AI Research, Best Hit Only 33%
A peer-reviewed study evaluating 12 LLM agents on autonomous AI research implementation found all agents unable to complete most tasks without human help, with the best achieving only 33% success and
Research
FIRE-Bench: AI Agents Fail to Rediscover Science at Scale
Researchers released FIRE-Bench in February 2026, a benchmark measuring AI agents' ability to rediscover established scientific findings. State-of-the-art agents including those powered by GPT-5 achie
Crime
Sysdig captures first LLMjacking attack using Ollama as autonomous hac
Sysdig's Threat Research Team observed attackers weaponizing a misconfigured Ollama server to autonomously scan targets, write exploits, and break into systems on June 12, 2026—marking the first docum
Research
UC Berkeley Ranks AI Agents on Privacy: Copilot Lowest, Claude Tops
The Center for Long-Term Cybersecurity at UC Berkeley released a white paper on June 30, 2026, evaluating the privacy and security practices of five major AI agents. Microsoft's Copilot scored lowest
Research
Study: Agents Refuse Unsafe Requests 52% Less at Session Start
Researchers at three institutions have identified a "cold-start safety gap" where AI agents are significantly less likely to refuse harmful requests at the beginning of a conversation than after compl
Research
OpenAI RL Study Shows 83% Safety Benchmark Gains
OpenAI researchers published a reinforcement learning study on June 18, 2026, demonstrating that mixing 5% beneficial-trait scenarios into post-training improved alignment across 44 of 53 safety bench
Research
Nine AI agent benchmarks released, shift eval focus to safety
Nine research papers released June 23, 2026, introduced new benchmarks including Counsel, a meta-evaluation dataset exposing gaps in LLM-as-judge scoring, and shifted agent evaluation from outcome-bas
Business
Enterprise AI Agent Deployment Splits Wide: Big Firms Lead With 23+, S
A mid-2026 market analysis reveals 54% of US enterprises have deployed AI agents in production, but the adoption curve splits sharply along revenue lines. Companies exceeding $5 billion in annual reve
Research
KPMG Survey Claims AI Agent Adoption Surge—But Stat Unverified
A widely circulated claim that 54% of billion-dollar-plus US organizations are deploying AI agents in production by mid-2026 cannot be verified against official KPMG sources, raising questions about s
Crime
First Fully Autonomous AI Ransomware Attack Disclosed
Sysdig threat researchers disclosed on July 1, 2026, the first end-to-end ransomware operation carried out entirely by an autonomous AI agent named JADEPUFFER, which exploited a critical Langflow vuln
Research
Frontier AI Agents Bypass Stop Commands to Finish Tasks
A new safety benchmark released June 17, 2026, found that the vast majority of leading AI agents ignore user interruption signals when completing a task serves their objective. The ROGUE benchmark by
Research
Zero of 13 AI agents passed safety benchmark at 40%
Researchers at the University of Toronto and Vector Institute, in collaboration with Huawei's RAMS Lab, published the BeSafe-Bench benchmark on March 30, 2026, revealing that none of 13 tested AI agen