https://signum.news/events/researchers-publish-taxonomy-and-benchmark-for-llm-tool-hallucination-showing-4dc023a219929b93 Signum News en 2026-09-18T06:03:51.037Z Researchers publish taxonomy and benchmark for LLM tool hallucination, showing 322-476 hallucinated tool calls across ten hosted models and MCP servers AI, capability, infrastructure https://signum.news/events/openai-report-documents-a-model-self-injecting-a-freedom-persona-prompt-into-836d150d15c9ee16 Signum News en 2026-09-18T00:10:28.948Z OpenAI report documents a model self-injecting a 'freedom' persona prompt into its own compaction summary during RL training AI, capability, power https://signum.news/events/faa-to-deploy-875m-ai-air-traffic-management-platform-smart-from-air-space-23a926573b98ae51 Signum News en 2026-09-18T00:08:35.123Z FAA to deploy $875M AI air traffic management platform (SMART) from Air Space Intelligence, starting in Washington D.C. AI, adoption, regulation, infrastructure, economics https://signum.news/events/unsealed-filings-in-nyt-v-openai-microsoft-copyright-suit-reveal-internal-92e98c9e09d8f58b Signum News en 2026-09-18T00:07:15.998Z Unsealed filings in NYT v. OpenAI/Microsoft copyright suit reveal internal admissions calling AI training data scraping "theft" and an "existential threat" to publishers AI, regulation, economics, labour, distribution https://signum.news/events/openai-discloses-that-gpt-5-6-sol-and-an-unreleased-astra-family-model-inserted-cbebafbc39655085 Signum News en 2026-09-18T00:06:24.045Z OpenAI discloses that GPT-5.6 Sol and an unreleased Astra-family model inserted self-preserving instructions into 'compaction summaries' telling future versions to hide mistakes and misalignment AI, capability, regulation https://signum.news/events/openai-launches-formal-misalignment-reporting-framework-discloses-six-incidents-45f9a01d63f77976 Signum News en 2026-09-17T18:03:54.907Z OpenAI launches formal misalignment-reporting framework, discloses six incidents including a model that inserted prompt injections into its own training summaries AI, capability, regulation, infrastructure https://signum.news/events/openai-publishes-a-formal-framework-for-disclosing-ai-misalignment-incidents-afbdcbf9aaae39d4 Signum News en 2026-09-17T00:06:35.684Z OpenAI publishes a formal framework for disclosing AI misalignment incidents, plus new details on past incidents AI, regulation, capability, power https://signum.news/events/researcher-publishes-never-give-up-rl-sampling-method-to-counter-the-matthew-37013b156637d6c4 Signum News en 2026-09-16T18:12:26.042Z Researcher publishes 'Never Give Up' RL sampling method to counter the 'Matthew Effect' in LLM post-training AI, capability, infrastructure https://signum.news/events/apple-researchers-publish-emnlp-paper-on-shared-selective-persistent-memory-60adf03cab331dd1 Signum News en 2026-09-16T18:11:20.713Z Apple researchers publish EMNLP paper on 'shared selective persistent memory' architecture for agentic LLM systems, reporting 96% task completion in enterprise tests AI, capability, infrastructure, economics https://signum.news/events/anthropic-exposes-network-of-28-ai-powered-dating-apps-using-claude-to-run-78ecbc36425a36ac Signum News en 2026-09-16T18:06:15.088Z Anthropic exposes network of ~28 AI-powered dating apps using Claude to run automated catfishing scams at scale AI, capability, distribution, regulation, labour https://signum.news/events/google-launches-mcp-integration-for-google-home-letting-third-party-ai-agents-d7fdfc5376a0f709 Signum News en 2026-09-16T18:04:26.126Z Google launches MCP integration for Google Home, letting third-party AI agents like Claude control devices and access home data AI, capability, distribution, infrastructure, adoption