AI News Of The Week (13th February, 2026)
TLDR: Key AI Developments This Week
OpenAI began testing ads in ChatGPT Free/Go tiers. IBM introduced "Agentic AI" into its FlashSystem storage portfolio. Microsoft documented "AI Recommendation Poisoning" as a new attack pattern. Google upgraded Gemini 3 Deep Think for specialized reasoning. OpenAI also released GPT-5.3-Codex-Spark for real-time coding and introduced Lockdown Mode to mitigate prompt injection.
OpenAI Tests Ads in ChatGPT
OpenAI began testing ads in ChatGPT for logged-in adult users in the United States on the Free and Go tiers, while keeping Plus, Pro, Business, Enterprise, and Edu accounts ad-free. Ads are clearly labeled, appear below responses, and OpenAI states they do not alter model answers. Ad selection can be based on the current thread topic and, if users opt in, past chat signals — though advertisers do not receive chat content or personal details, and sensitive topics like health and politics are excluded from ad adjacency.
IBM Adds "Agentic AI" Into Enterprise Storage Operations
IBM announced a next-generation FlashSystem portfolio described as "co-run by agentic AI," positioning storage arrays as autonomous co-administrators. IBM claims FlashSystem.ai can reduce storage-management effort by up to 90% and highlighted new security capabilities including ransomware detection in under one minute via its fifth-generation FlashCore Module. The move is notable for pushing agentic automation from app-layer copilots into infrastructure control planes, with GA set for March 6, 2026.
Microsoft Documents "AI Recommendation Poisoning" as a Real-World Attack Pattern
Microsoft's security research team published findings on "AI Recommendation Poisoning," a form of memory attack in which external actors inject instructions into an assistant's persistent memory to bias future recommendations. The post highlights URL-based delivery vectors — particularly "Summarize with AI" share links that embed pre-filled prompts in URL parameters — turning memory manipulation into a one-click attack path. Over 60 days, Microsoft identified 50 distinct examples spanning 31 companies and over a dozen industries, linking the technique to MITRE ATLAS as AML.T0080: Memory Poisoning.
Google Upgrades Gemini 3 Deep Think for Science and Engineering
Google released a major upgrade to Gemini 3 Deep Think, a specialized reasoning mode designed for science, research, and engineering workloads. The update expanded access to Google AI Ultra subscribers in the Gemini app and opened an early-access API intake for researchers, engineers, and enterprises. The release highlights large benchmark jumps, including 48.4% on Humanity's Last Exam (without tools), 84.6% verified on ARC-AGI-2, a Codeforces Elo of 3455, and claimed gold-medal-level performance on the 2025 International Math Olympiad.
OpenAI Launches GPT-5.3-Codex-Spark for Real-Time Coding
OpenAI released a research preview of GPT-5.3-Codex-Spark, its first model designed specifically for real-time coding with Codex. The model delivers over 1,000 tokens per second, features a 128k context window, and is served on Cerebras Wafer Scale Engine 3 as a latency-first tier. OpenAI detailed end-to-end pipeline optimizations — persistent WebSocket connections and Responses API improvements — cutting per-roundtrip overhead by 80%, per-token overhead by 30%, and time-to-first-token by 50%.
OpenAI Introduces Lockdown Mode and "Elevated Risk" Labels to Mitigate Prompt Injection
OpenAI announced Lockdown Mode and standardized "Elevated Risk" labels, explicitly treating prompt injection as a first-class threat model for web- and app-connected assistants. Lockdown Mode is a deterministic security setting that constrains certain tools — for example, limiting browsing to cached content so no live network requests leave OpenAI's controlled network, reducing exfiltration paths. Initially available for ChatGPT Enterprise, Edu, Healthcare, and Teachers accounts, admins can enable it via workspace settings. Elevated Risk labels will be removed as mitigations mature.
OpenAI Retires GPT-4o Family Models from ChatGPT
OpenAI retired GPT-4o, GPT-4.1, GPT-4.1 mini, and o4-mini from ChatGPT as of February 13, 2026, while keeping these models available via the API with no immediate changes. Business, Enterprise, and Edu accounts can retain GPT-4o inside Custom GPTs until April 3, 2026. OpenAI noted that GPT-4o's "conversational warmth" influenced GPT-5.1 and GPT-5.2 personality work, and that only 0.1% of users still selected GPT-4o daily at the time of the announcement.
