AI Safety news and analysis on Web Pulse: 12623 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
EU Energy Commissioner Dan Jorgensen warned Big Tech that AI data centres are welcome in Europe only if they align with the bloc's climate goals, including using renewable or nuclear power and recycling waste heat. The w…
Nobel Prize-winning AI pioneer Geoffrey Hinton criticized Elon Musk's xAI and warned of growing risks in AI development, declining an advisory role at the company. Hinton raised concerns about job displacement, safety fa…
Swiss startup Prem AI is raising $100 million in Series A funding at a $500 million valuation to provide hedge funds and law firms with private, self-hosted AI infrastructure. The company launched an encrypted AI workspa…
A developer argues that AI will replace 'I-shaped' specialists but not 'T-shaped' or 'comb-shaped' generalists who can adapt and understand context. The full-stack engineer, as a generalist, will survive the AI disruptio…
A malware developer is embedding text about nuclear and biological weapons in spyware to deter AI analysis. The malicious code begins with a JavaScript block comment containing fake system instructions and policy-trigger…
Prem AI, a developer of private AI infrastructure for sovereign data solutions, is raising $100 million in a Series A funding round, according to Bloomberg. The round would be a massive leap for the company, which has di…
A growing reliance on AI-generated code is creating a dangerous feedback loop where AI systems generate and review code with minimal human oversight, leading to increased defects and security vulnerabilities. According t…
Cisco Talos published a technical note demonstrating how the VB6 disassembler vbdec can enable local, agent-driven reverse engineering by exposing its parsed data model through the Windows Running Object Table. The techn…
A developer outlines a security model for AI agents that access employee data, emphasizing three principles: separating read and write agents, generating immutable audit records for every query, and scoping inference to …
A new study from researchers at multiple labs found that adversarial manipulation of external information feeds can steer LLM agents' decisions away from their defaults, with effects ranging from 5% to 100% in some cases…
A controlled experiment sent 1,000 fake visitors to a site running Google Analytics 4 and Clickport. GA4 flagged none of the bot sessions, while Clickport blocked 80% across five scenarios. The test highlights GA4's fail…
Honrly, a local-first, self-hostable interview-integrity tool, has released its source code, demonstrating that AI cheating tools like Cluely and Interview Coder are easily detectable. The tool scans running processes in…
A developer built Sego, a local-first review layer for AI-generated code that sits after coding agents to assess safety before committing. The MVP reviews staged changes and produces structured findings with severity, ev…
OpenAI has expanded ChatGPT's memory to build a profile from past chats, including entire chat history and implicit preferences, but old or irrelevant details can distort future AI answers. The feature, called 'dreaming,…
Maref, a multi-agent recursive engineering framework, launched as the first product focused on agent governance, featuring a 10-state Gray code state machine, TLA+ formal verification, and a four-level safety decision tr…
A developer built NPM Safety Guard, a free IDE extension that scans package.json and lockfiles for malicious npm packages across 22 detection layers. The tool catches typosquats, AI-hallucinated package names, suspicious…
Israeli AI and cybersecurity firm Dream raised $260 million at a $3 billion valuation, nearly tripling its $1.1 billion valuation from 16 months ago. The round was led by Bicycle Capital and Group 11, with participation …
Microsoft's VS Code 1.123, released June 3, introduces a two-hour delay before newly published extension versions auto-update, aiming to mitigate supply chain attacks by providing a window to catch malicious updates. The…
Tinfoil (YC X25) proposes verifiable privacy for cloud AI inference pipelines using cryptographic proofs, but the implementation gap lies in inspecting model artifacts before deployment. Local model files like .gguf and …
A web developer nearly fell victim to a sophisticated phishing attack disguised as a Calendly booking link from a potential client, HitBox Games. The attacker used a fake browser window mimicking Google's login page to s…