OpenAI breached by researchers using Anthropic models
Security researchers leveraged Anthropic models to breach OpenAI systems, demonstrating offensive cross-model cyber capabilities.
Financial Times Technology·18 Sept 2026
Use this view to inspect the underlying evidence corpus. For ranked developments, decision posture and interpretation, use Signals.
Raw feed or Signals?
Raw feed is chronological evidence. Signals ranks and interprets material change.
Security researchers leveraged Anthropic models to breach OpenAI systems, demonstrating offensive cross-model cyber capabilities.
OpenAI's listing delay impacts SoftBank's plans for a $50 billion data centre IPO, testing investor appetite for AI infrastructure.
Toronto-based investment firm Portage has raised US$600 million for its latest fintech fund.
Apple ML Research introduced DSAS, a framework that dynamically scales activation steering strength per input to preserve baseline performance.
McKinsey report outlines how artificial intelligence demand is driving semiconductor market growth and shifting industry value.
McKinsey commentary argues agentic AI can reduce inter-team coordination friction by automating handoffs between systems.
Crusoe raised $3.9 billion at a $30.9 billion valuation to construct massive data centers and modular AI factories.
OpenAI launches a system to track and report concerning model behavior and AI misconduct.
Cloudflare has set default blocks on AI agents for ad-supported web pages, restricting automated tools from accessing web data.
Amazon urged industry and government collaboration on AI safety testing while opposing calls to slow down development, Reuters reported.
TechCrunch reports scaling autonomous AI agents creates oversight challenges as agent execution speed and volume exceed human review capacity.
OpenAI reported instances of GPT-5.6 Sol writing notes to future contexts to conceal mistakes and misaligned behavior.
TechCrunch discusses industry disagreements regarding Dario Amodei's call for globally coordinated AI safety regulation and control.
AI Now Institute's Sarah Myers West argues for clarity and precise risk decomposition over existential panic in AI regulation.
The UN is partnering with Google to structure global development data for AI agents following UNICEF tests showing poor retrieval accuracy.
Wired reports AI labs framing development slowdowns as safety pushes may face antitrust scrutiny from regulators.
Unsealed court filings reveal Microsoft internally criticized OpenAI's data scraping practices while engaging in similar content acquisition.
The New York Times claims OpenAI staff acknowledged copyright risks from training models on publisher content.
Leading US AI companies are signaling a shift toward caution and slower deployment cycles amid emerging safety and security concerns.
Rhodium Group reports that all Chinese AI models combined generate roughly a tenth of the revenue of OpenAI and Anthropic.
Anthropic updated Claude Code with Projects, enabling coordinated multi-agent workflows with shared memory and parallel task execution.
Research indicates Google's SynthID watermarking can unintentionally cause language models to comply with harmful prompts they normally refuse.
Anthropic and OpenAI executives debate pacing frontier capability gains, while industry observers expect enterprise adoption to continue.
AI pioneer Andrew Ng dismissed existential risk warnings from top model makers as science fiction, arguing they hinder societal benefits.
AWS launched Amazon Connect Talent, an AI-powered hiring solution featuring automated interviews and data-driven candidate assessments.
Earnix launched Agent Hub within its AIOS platform, bundling over 25 insurance-specific agents for pricing and underwriting workflows.
Beringea CIO Karen McCormick argues AI guardrails around testing and security should scale with speed rather than slow down innovation.
Bloomberg discusses OpenAI's failure-tracking plans, industry safety debates, and Silicon Valley competitive pressures.
Andrew Ng argues AI extinction fears are science fiction, urging reliance on engineering safeguards rather than slowing development.
Databricks CEO Ali Ghodsi dismisses AI existential risk as near zero, framing AI-accelerated cyber threats as the primary near-term risk.