17 February 2026

 

Here’s today’s concise AI + genealogy briefing for Tuesday, 17 February 2026.

---

1. Notable AI engine and tool updates

• OpenAI is rolling out Lockdown Mode and Elevated Risk labels in ChatGPT, Atlas, and Codex to give admins tighter control over risky features and reduce data‑exfiltration risks, with broader consumer rollout planned over the coming months.[releasebot]​

• Recent OpenAI platform updates improve memory handling stability and reduce noisy developer messages entering phase‑1 memory, aiming for more reliable long‑running or “agent‑style” sessions that recall prior context without drifting.[releasebot]​

• The same OpenAI release train adds an experimental JavaScript REPL runtime that can persist state across tool calls, plus better multi‑channel rate‑limit handling and a reintroduced WebSocket transport, all of which support faster, more interactive assistants and agents.[releasebot]​

• End‑to‑end latency optimizations in OpenAI’s inference stack now shorten time‑to‑first‑token and keep Codex‑style models responsive for real‑time collaboration, which indirectly benefits other models using the same serving pipeline.[releasebot]​

• OpenAI is also improving a “deep research” mode in ChatGPT so users can constrain research to specific websites or trusted app connections, plan research runs in advance, and adjust direction mid‑run via a redesigned fullscreen report interface.[releasebot]​

• OpenAI has scheduled the retirement of older ChatGPT model snapshots such as chatgpt‑4o‑latest from the API today (17 February 2026), with migrations steering usage toward newer GPT‑5.x family models.[developers.openai]​

• Anthropic’s latest Claude platform releases tighten safeguards against nesting Claude Code sessions, improve streaming, session previews, and error messages, and fix crashes tied to image‑returning tools, making Claude‑based coding and agent workflows more stable.[releasebot]​

• The Claude developer tooling now includes CLI authentication helpers, Windows ARM64 support, and better handling of Bedrock, Vertex, and Foundry model identifiers, aimed at smoothing enterprise and multi‑cloud deployments.releasebot+1

• Anthropic’s Claude Opus 4.6, launched earlier this month, is positioned as its most capable flagship with improved coding skills and large‑context “agent teams,” while older Opus 4 and 4.1 models are deprecated in the console and Claude Code.anthropic+1

• Wider ecosystem news this week highlights OpenAI’s GPT‑5.3 Codex Spark (a high‑speed, Cerebras‑hosted coding model), Anthropic’s new funding at a reported 380‑billion‑dollar valuation, Alibaba’s open‑weight Qwen3.5 model, and increasing emphasis on autonomous “do‑things” agents like OpenClaw.[aijungle.substack]​

• OpenAI has also been updating ChatGPT’s default GPTs to use GPT‑5.2 as of mid‑February, so most users invoking generic “ChatGPT” experiences are now riding on that model unless they select something custom.[help.openai]​

---

2. Twenty-plus concrete AI use cases for genealogists

Each item is something a working genealogist or family‑history blogger can try today with a capable chat model, always verifying against original records.

1. Turn raw research notes into a research log
   Paste yesterday’s scattered notes (citations, URLs, archive call numbers) and ask the model to restructure them into a table with columns like source, jurisdiction, date range, and “next action,” ready to paste into Excel or a research log template.[perplexity]​

2. Abstract deeds and land records
   Feed a deed transcript and ask for a neutral abstract listing grantor, grantee, consideration, metes and bounds, neighbors, and prior deed references, explicitly instructing the model not to infer relationships or locations not stated in the text.[perplexity]​

3. Generate research questions from a family group sheet
   Provide a small family group with dates and places, then ask the AI to list unanswered questions (missing parents, unexplained gaps, migration questions) plus specific record types and jurisdictions to investigate for each question.[perplexity]​

4. Help design reasonably exhaustive searches
   Describe a stubborn identity problem and what you have already checked, then ask for a structured plan that enumerates additional record sets (local, county, state, federal, and private) and time frames, which you can refine against catalogues like FamilySearch and ArchiveGrid.[perplexity]​

5. Summarize long DNA‑correspondence threads
   Paste a long email chain with a DNA match and ask for a concise summary of hypothesized relationships, key pieces of evidence, and agreed next steps, taking care not to reveal living‑person details beyond what you provide.[perplexity]​

6. Draft plain‑language explanations for non‑genealogist relatives
   Give the model a technical proof argument and ask it to produce a short, non‑technical explanation for cousins—explaining why you believe two profiles are the same person, what sources were used, and what uncertainties remain.[perplexity]​

7. Outline a blog post series from a case study
   Paste an existing report or proof argument and ask for a four‑part blog‑post outline that preserves your reasoning but breaks it into reader‑friendly episodes (background, evidence, analysis, resolution), with suggested subheadings.[perplexity]​

8. Create alternative “house‑style” drafts of a narrative
   Provide a paragraph or two of ancestor narrative and ask for a revision that follows your preferred “house rules” (no invented dialog, dates in a specific format, citations left as bracketed placeholders, no speculation beyond what you mark as such).[perplexity]​

9. Build timelines that correlate evidence across sources
   Paste extracts from multiple records for one person or family and ask the model to lay out a dated timeline with columns for event, source, informant (if known), and reliability notes, flagging conflicts for you to resolve.[perplexity]​

10. Generate alternative hypotheses for identity problems
    Describe an identity or same‑name problem and current working hypothesis, then ask the AI to list other plausible hypotheses, what each would predict in the record trail, and which tests (record searches, cluster analysis) could discriminate between them.[perplexity]​

11. Cluster‑research helper prompts
    Provide a small FAN‑club list (witnesses, neighbors, sponsors) and ask for ideas on how to prioritize them—for example, which surnames or occupations recur, who appears in multiple jurisdictions, and which look like strong migration companions to investigate first.[perplexity]​

12. Translate and preliminarily parse foreign‑language records
    Paste a short record in a language you partially read (for example, French, German, Dutch, Swedish) and ask for a literal translation, identification of key genealogical elements (names, relationships, places, dates), and a glossary of recurring terms and abbreviations; then compare carefully to the original.[perplexity]​

13. Draft repository visit plans
    Describe an upcoming trip to a specific repository (county courthouse, state archive, local library), including your research goals, and ask the model to help you prioritize record groups, script specific call‑slip requests, and draft a one‑page pull‑list.[perplexity]​

14. Suggest metadata and tags for digital filing systems
    Paste a list of file names or brief descriptions of digitized documents, and ask for proposed folder names and tags organized by surname, place, record type, and time period, which you can adapt to your own digital‑organization scheme.

15. Clean up citations at scale (with caution)
    Feed a batch of messy draft citations and ask the AI to normalize them toward a style you specify (for example, Evidence Explained‑inspired structure), while you spot‑check each transformed citation against your style guide and the underlying source.[perplexity]​

16. Create teaching handouts for society presentations
    Give the outline of an upcoming talk (for example, on land records or DNA evidence) and ask for a one‑page handout draft with key terms, example questions, and space for attendees’ notes, leaving placeholders where you will later insert your own real‑world examples.[perplexity]​

17. Summarize long probate files or case bundles
    Paste or upload transcripts of a long probate packet or court case and ask for a structured summary listing heirs, relationships as explicitly stated, property categories, and chronological case milestones, while you manually verify any relationship labels.[perplexity]​

18. Brainstorm titles and subheads for blog posts
    Provide a finished draft and ask for ten possible titles and several subhead options that emphasize evidence, methodology, or locality, then choose or adapt one that best fits your blog’s tone.

19. Convert research reports into slide decks
    Paste a section of a report and ask the AI to suggest a slide‑by‑slide outline (one idea per slide, brief bullets, suggested visuals like maps or record snippets) suitable for PowerPoint or other presentation tools.

20. Design controlled experiments comparing AI and traditional workflows
    Describe one ongoing ancestor problem and ask the model to help you outline an experiment: which subtasks you’ll attempt with AI (such as abstracting, outlining, translation), how you’ll measure time saved or added errors, and how you might later write this up for your blog or a society newsletter.History+1

21. Draft FAQs for your website or client materials
    List common questions relatives, readers, or clients ask (about online trees, conflicting dates, or DNA surprises) and ask the AI to draft short, evidence‑focused FAQ answers you will fact‑check and adapt to your voice before publishing.[perplexity]​

22. Maintain a “do‑over” or methods journal with AI’s help
    After each working session, dictate a quick summary of what you tried, where AI helped or hindered, and what you’ll do next, then ask the model to turn that into a dated journal entry you can later mine for teaching examples or reflective posts.[perplexity]​

You can treat these as a menu: pick one active ancestor file or blog draft that is already on your desk, plug it into one or two of these patterns (for example, deeds plus timeline, or translation plus narrative), and then test carefully where AI genuinely saves time versus where traditional close reading remains essential.History+1