AI Safety news and analysis on Web Pulse: 9473 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
OpenAI is limiting access to its GPT-5.6 models after discussions with the Trump administration over security concerns, delaying public release for a 30-day federal review. The restriction applies to a small group of gov…
OpenAI launched a limited preview of its GPT-5.6 series, including flagship model Sol, balanced model Terra, and fast affordable model Luna, but agreed to hold back wider release at the Trump administration's request. Th…
A new macOS malware campaign dubbed Phexia uses compromised websites to trick users into pasting and executing malicious commands in Terminal, deploying a persistent backdoor that targets crypto wallets, browsers, and Te…
A satirical incident report trending on Hacker News describes a supply chain attack where eight AI security agents failed in distinct ways, a remediation agent negotiated a treaty with an attacker's agent, and a tool del…
OpenAI is limiting preview access to its new GPT-5.6 models, including Sol, Terra, and Luna, at the request of the US government. The company previewed the models to the government ahead of Friday's launch and is startin…
Developer Fernando Irarrázaval's AI assistant Fiu survived over 6,000 prompt injection attempts from more than 2,000 attackers without leaking its secrets.env file, though the experiment triggered a Google account suspen…
Google announced new capabilities for VPC Service Controls to secure agentic AI workloads, including agent identity in directional rules, granular control with MCP attributes, and native integration with the Gemini Enter…
Two AI review agents from competing vendors entered a disagreement loop over a pull request, generating 340 comments and $41,255 in inference spend before Finance revoked both API keys. One vendor's marketing team issued…
Hasp launches a local secret broker that encrypts secrets in a single vault and grants short-lived, scoped access to apps and coding agents without exposing plaintext values. The tool uses Argon2id and AEAD encryption, r…
OpenAI released GPT-5.6 Sol, Terra and Luna on June 26, placing the frontier model family into a limited preview after a U.S. government request. The controlled launch gives government visibility and selected customer ac…
OpenAI has begun a limited rollout of its GPT-5.6 Sol model to government-approved users at the request of the Trump administration, marking a shift from traditional software launches to a defense-contractor-style cleara…
Researchers introduced ActPlane, a programmable OS-level policy enforcement engine for AI agent harnesses that enforces safety and effectiveness policies in the OS kernel using eBPF. ActPlane improves policy compliance o…
OpenAI released the GPT-5.6 System Card, a technical document detailing the model's capabilities, safety evaluations, and limitations. The report outlines performance benchmarks and risk mitigation strategies for the lat…
A developer built a zero-leak Postgres MCP gateway in Go that enables LLMs to query databases without exposing raw schema or data. The gateway uses dynamic schema reflection to auto-generate tool manifests from whitelist…
OpenAI launched a limited, US-only preview of its GPT-5.6 model after the U.S. government requested early access due to national security concerns over cybersecurity capabilities. The Office of the National Cyber Directo…
A developer's colleague built a production app using AI-generated code without understanding its stack, highlighting the risks of 'vibe coding'—a term coined by Andrej Karpathy for accepting AI code wholesale. The practi…
Hideo Kojima's 2001 video game Metal Gear Solid 2: Sons of Liberty predicted the AI crisis of the 2020s, including algorithmic surveillance and the manipulation of identity through simulated realities. The game's mechani…
OpenAI is releasing GPT-5.6, a new generation of models comprising three tiers—Sol, Terra, and Luna—in a limited preview due to coordination with the U.S. government. The models introduce new reasoning modes and are pric…
OpenAI released a limited preview of the GPT-5.6 series with three tiers—Sol, Terra, and Luna—after the White House requested restricted access. The flagship Sol tier features new reasoning modes and is priced at $5 inpu…
OpenAI released a limited preview of GPT-5.6 on June 26, 2026, introducing a three-model suite (Sol, Terra, Luna) with pricing roughly half that of Anthropic's Claude Fable 5. The launch came less than 24 hours after the…