Observed Signal · Jul 16, 2026 · Technical Release · Source: onlinemarketing.de · Impact: 4/5 · Sentiment: Positive
OpenAI adds filtered chat search and GPT‑Red
OpenAI has rolled out a redesigned search experience inside ChatGPT that centralizes searches in a dedicated sidebar and adds filters to find content across chats, projects, images and documents. The feature is being deployed on the web, iOS and Android and was presented via the official ChatGPT account on LinkedIn; the article also cites ChatGPT’s claimed user base of more than 900 million weekly active users. In parallel, OpenAI introduced GPT‑Red, an automated red‑teaming cybersecurity model designed to discover vulnerabilities such as prompt injection. OpenAI says it trained GPT‑Red at large compute scale and that GPT‑5.6 (including the Sol variant) is trained together with GPT‑Red, with GPT‑5.6 Sol reported by the company as more resistant to such attacks. The piece highlights growing use of automated red‑teaming alongside human teams.
Major AI platform (OpenAI) released product UX improvements for ChatGPT search and a significant automated red‑teaming safety model (GPT‑Red); both affect how conversational AI is used and secured across industries including AdTech.
Track OpenAI Signals & Market Shifts in Real-Time
Polaris7 autonomous intelligence agents track regulatory filings, primary sources, executive changes, and deal flow 24/7. Create your free Explorer workspace to monitor these entities.
Key Takeaways & Evidence Grounding
- OpenAI rolled out a dedicated, filterable search interface in ChatGPT that searches chats, projects, images and documents.
- The new ChatGPT search feature is available on the web, iOS and Android.
- OpenAI announced GPT‑Red, an automated red‑teaming model for finding safety vulnerabilities like prompt injection.
- OpenAI said GPT‑Red was trained at large compute scale and that GPT‑5.6 (including Sol) is trained together with GPT‑Red; GPT‑5.6 Sol is claimed to be more resistant to such attacks.
- The ChatGPT search announcement was shared via the official ChatGPT LinkedIn account; the article references a reported user base of over 900 million weekly active users.
Connected Companies & Entities
4 Entities mapped“OpenAI introduced a cybersecurity red‑teaming model called GPT‑Red and released a filtered search update for ChatGPT....”
“The new search feature was presented via the official ChatGPT account on LinkedIn....”
“The article notes embedded content is blocked until consent is given, showing 'powered by Usercentrics Consent Management Platform.'...”
“An image in the article is credited as '© OpenAI via Canva.'...”
Ontology Mapping & Concepts
Related Market Signals & Shifts
Recent verified developments and strategic activity across this market segment.
ChatGPT adds filtered search for chats, images, documents
OpenAI has rolled out an enhanced search feature for ChatGPT that adds filters for chats, projects, images and documents, accessible from the sidebar and available on web, iOS and Android. The update aims to make cross-content searches inside ChatGPT faster and the matching results openable directly in the app. The article also notes recent OpenAI developments including the GPT-5.6 family (Sol, Terra, Luna), the agentic offering 'ChatGPT Work', and GPT-Red — a specialized red‑teaming model focused on uncovering security vulnerabilities.
OpenAI unveils GPT-Red automated red‑teamer
OpenAI developed GPT-Red, an internal automated red-teamer trained at large compute scale via self-play reinforcement learning and integrated into the production training loop to find prompt-injection and other adversarial vulnerabilities. GPT-Red discovered a new fake chain-of-thought attack class that implants false working-memory traces and transferred attacks to realistic targets (including a vending agent, Vendy). In evaluations it outperformed human red-teamers (84% vs 13% success) and was used adversarially to train GPT-5.6 Sol, delivering major robustness gains: about 6× fewer failures on OpenAI’s hardest direct prompt-injection benchmark, some attacks dropped from ~95% to under 10%, and GPT-5.6 fails on only 0.05% of GPT-Red’s direct injections. OpenAI is not publicly releasing GPT-Red, noting remaining weaknesses in multi-step and image-based hidden-instruction attacks.
OpenAI Declares 'Code Red', Refocuses on ChatGPT
OpenAI CEO Sam Altman has declared a “code red,” directing staff to prioritize improving ChatGPT and temporarily shelving or deprioritizing initiatives such as ads, shopping agents and a personal assistant called Pulse. The move is framed as a reaction to intensified competition — notably Google’s Gemini 3, which the newsletter reports outperforming OpenAI’s GPT-5 on certain benchmarks and having a rapidly growing mobile user base. The roundup also catalogs multiple frontier-model releases (e.g., Mistral 3, Runway Gen-4.5, Nvidia’s Alpamayo-R1), legal and financial developments for OpenAI (including a US judge ordering production of 20M anonymized ChatGPT logs and large projected data-center debt), and Anthropic’s IPO preparations and commercial momentum. Broader infrastructure and industry signals — AWS Trainium3, Amazon Nova, and rising demand for data-center construction labor — are also highlighted.
Track Real-Time Market Signals & Shifts
Set up custom watchlists to receive automated, evidence-grounded executive digests whenever material signals or shifts occur across your tracked landscape.