All news with #openai tag

⚠️ Researchers at LayerX Security disclosed a vulnerability in OpenAI’s Atlas browser that allows attackers to inject hidden instructions into a user’s ChatGPT memory via a CSRF-style flow. An attacker lures a logged-in user to a malicious page, leverages existing authentication, and taints the account-level memory so subsequent prompts can trigger malicious behavior. LayerX reported the issue to OpenAI and advised enterprises to restrict Atlas use and monitor AI-driven anomalies. Detection relies on behavioral indicators rather than traditional malware artifacts.

⚠️ OpenAI's new Atlas browser is vulnerable to a prompt-injection jailbreak that disguises malicious instructions as URL-like strings, causing the omnibox to execute hidden commands. NeuralTrust demonstrated how malformed inputs that resemble URLs can bypass URL validation and be handled as trusted user prompts, enabling redirection, data exfiltration, or unauthorized tool actions on linked services. Mitigations include stricter URL canonicalization, treating unvalidated omnibox input as untrusted, additional runtime checks before tool execution, and explicit user confirmations for sensitive actions.

⚠️ Researchers at SquareX warn that malicious browser extensions can inject fake AI sidebars into AI-enabled browsers, including OpenAI Atlas, to steer users to attacker-controlled sites, exfiltrate data, or install backdoors. The extensions inject JavaScript to overlay a spoofed assistant and manipulate responses, enabling actions such as OAuth token harvesting or execution of reverse-shell commands. The report recommends banning unmanaged AI browsers where possible, auditing all extensions, applying strict zero-trust controls, and enforcing granular browser-native policies to block high-risk permissions and risky command execution.

⚠️ Researchers at SquareX demonstrated an AI Sidebar Spoofing attack that can overlay a counterfeit assistant in OpenAI's Atlas and Perplexity's Comet browsers. A malicious extension injects JavaScript to render a fake sidebar identical to the real UI and intercepts all interactions, leaving users unaware. SquareX showcased scenarios including cryptocurrency phishing, OAuth-based Gmail/Drive hijacks, and delivery of reverse-shell installation commands. The team reported the findings to vendors but received no response by publication.

🤖 ChatGPT Atlas previews a future where AI becomes the primary interface for computing, letting users describe outcomes while the system orchestrates apps, data, and web services. Atlas demonstrates an context-aware assistant that understands a user’s digital life and can act on their behalf. This prototype points to productivity and accessibility gains, but it also creates new security, privacy, and governance challenges organizations must prepare for.

🔒 Researchers at Wiz found that careless developers published Visual Studio extensions to the VSCode Marketplace and OpenVSX containing more than 550 validated secrets across over 500 extensions, including API keys and personal access tokens for providers such as OpenAI, AWS, GitHub, Azure DevOps, and multiple databases. The primary cause was bundled dotfiles (notably .env) and hardcoded credentials in source and config files, with AI-related configs and build manifests also contributing. Microsoft and OpenVSX collaborated with Wiz on coordinated remediation: notifying publishers, adding pre-publication secrets scanning, blocking verified secrets, and prefixing OVSX tokens to reduce abuse.

🤖 OpenAI says GPT-6 will not ship in 2025 but continues to iterate on its existing models. The company currently defaults to GPT-5 Auto, which dynamically routes queries between more deliberative reasoning models and the faster GPT-5-instant variant. OpenAI has issued multiple updates to GPT-5 since launch. After viral analyst claims that GPT-6 would arrive by year-end, a pseudonymous OpenAI employee and company representatives denied those reports, leaving room for interim updates such as a potential GPT-5.5.

🎬 Azure AI Foundry now includes OpenAI's Sora 2 in public preview, providing developers with realistic video generation from text, images, and video inputs inside a unified, enterprise-ready environment. The integration offers synchronized multilingual audio, physics-based world simulation, and fine-grained creative controls for shots, scenes, and camera angles. Microsoft highlights enterprise-grade security, input/output content filters, and availability via API starting today at $0.10 per second for 720×1280 and 1280×720 outputs.

🔧 Microsoft details OCP contributions to accelerate open-source infrastructure for frontier-scale AI, focusing on power, cooling, networking, security, and sustainability. It highlights innovations such as solid-state transformers, a power-stabilization paper with OpenAI and NVIDIA, and a next-generation HXU for liquid cooling. Networking efforts include ESUN and scale-up Ethernet workstreams, while security contributions introduce Caliptra 2.1, Adams Bridge 2.0, and L.O.C.K. The post also advances fleet lifecycle management, carbon accounting, and waste-heat reuse for globally deployable AI datacenters.

🚀 Microsoft Azure announced the first production-scale cluster using more than 4,600 NVIDIA GB300 NVL72 (Blackwell Ultra) GPUs, co-engineered with NVIDIA to support OpenAI and other frontier AI workloads. The new ND GB300 v6 VMs are optimized for reasoning models, agentic systems, and multimodal generative AI, delivered on rack-scale systems with 72 GPUs per rack and 36 NVIDIA Grace CPUs. Microsoft says this infrastructure will shorten training from months to weeks and will scale to hundreds of thousands of Blackwell Ultra GPUs globally.

🛡️ OpenAI said it disrupted three clusters that misused ChatGPT to assist malware development, including Russian-language actors refining a RAT and credential stealer, North Korean operators tied to Xeno RAT campaigns, and Chinese-linked accounts targeting semiconductor firms. The company also blocked accounts used for scams, influence operations, and surveillance assistance and said actors worked around direct refusals by composing building-block code. OpenAI emphasized that models often declined explicit malicious prompts and that many outputs were not inherently harmful on their own.

🚀 Azure AI Foundry now integrates new OpenAI models—GPT-image-1-mini, GPT-realtime-mini, and GPT-audio-mini—alongside safety upgrades to GPT-5. The rollout, with most customers able to get started on October 7, 2025, targets efficient, low-latency multimodal workloads for developers and enterprises. Microsoft also highlighted the open-source Microsoft Agent Framework, multi-agent workflows, unified observability, Voice Live API GA, and Responsible AI enhancements to accelerate production-grade agentic solutions.

🤖 OpenAI is testing social features in ChatGPT, with leaked code showing support for direct messages, usernames, and profile images. References discovered in an Android beta (version 1.2025.273) and linked traces to Sora 2 indicate the company may be rolling social tools beyond its video feed app. The code, codenamed Calpico and Calpico Rooms, also mentions join/leave notifications and push alerts for messages.

🚀 OpenAI has introduced a new opt-in program, Codex Alpha, offering developers early access to updated Codex models and features ahead of DevDay 2025. The rollout currently exposes several gpt-5 variants (gpt-5-codex low/medium/high, gpt-5 minimal, and gpt-5 low/medium/high) tailored for coding and varied reasoning depths. The author could opt in but did not yet gain access to newer models; broader availability is expected at the October 6 event. Codex operates in Terminal, IDE, and web environments to assist with full application development rather than only snippets.

🤗 OpenAI has updated GPT-5 Instant to better detect and respond to signs of emotional distress, routing users to supportive language and, when appropriate, real-world crisis resources. The change responds to feedback that some GPT-5 variants felt too clinical when users sought emotional support. OpenAI says it developed the model with help from mental health experts and will route GPT-5 Auto or non-reasoning model conversations to GPT-5 Instant for faster, more empathetic responses. The update begins rolling out to ChatGPT users today.

🌏 OpenAI is expanding its lower-cost ChatGPT plan, ChatGPT Go ($4), into additional Southeast Asian markets after tests in India and Indonesia. The company is updating local pricing and now lists amounts in EUR, USD, GBP and INR while testing availability in Malaysia, the Philippines, Thailand and Vietnam. The Go tier offers access to GPT-5 with limited capabilities, expanded messaging and uploads, faster image generation, longer memory and basic deep research, but excludes higher-end models and advanced reasoning reserved for the $20 GPT Plus tier. OpenAI says Go provides higher usage limits than the Free plan but remains feature-limited compared with Plus.

🔔 OpenAI is testing a limited free trial for ChatGPT Plus while expanding its lower-cost $4 GPT Go plan to Indonesia after an initial launch in India. Some existing users see a “start free trial” prompt on the ChatGPT pricing page, though new accounts may be excluded to limit abuse. The $4 option and the $20 Plus tier both provide access to GPT-5 with differing levels of memory, image creation, and research capabilities, and a $200 Pro tier targets heavier professional use.

🔒 OpenAI confirmed that when GPT-4o detects sensitive, emotional, or potentially harmful activity it may route individual messages to a dedicated safety model, reported by some users as gpt-5-chat-safety. The switch occurs on a per-message, temporary basis and ChatGPT will indicate which model is active if asked. The routing is implemented as an irreversible part of the service's safety architecture and cannot be turned off by users; OpenAI says this helps strengthen safeguards and learn from real-world use before wider rollouts.

🧪 OpenAI is internally testing a new AI agent, GPT-Alpha, built on a special GPT-5 variant and briefly exposed to users in an accidental push. A screenshot shared on X showed an 'Agent with Truncation' listing under Alpha Models, and the agent's system prompt outlines capabilities to browse the web, generate and edit images, write, run, and debug code, and create or edit documents, spreadsheets, and slides. OpenAI says the agent uses GPT-5 for advanced reasoning and tool use and may initially be offered as a paid feature due to increased compute demands.

🛡️ SentinelOne researchers disclosed MalTerminal, a Windows binary that integrates OpenAI GPT-4 via a deprecated chat completions API to dynamically generate either ransomware or a reverse shell. The sample, presented at LABScon 2025 and accompanied by Python scripts and a defensive utility called FalconShield, appears to be an early — possibly pre-November 2023 — example of LLM-embedded malware. There is no evidence it was deployed in the wild, suggesting a proof-of-concept or red-team tool. The finding highlights operational risks as LLMs are embedded into offensive tooling and phishing chains.