Skip to main content
Skip to main content

Changelog

  1. Changelog

    Atlas now keeps the question, the corpus, and compare in one workspace.

    Research starts from official sources. Compare sits next to the corpus. The open question stays pinned while you read. Counsel decides. Software is not legal advice.

    Atlas names documented readings and evidenced carrying practice. It does not name which reading is legally true.

    #Research mode

    Research mode is built for counsel work on a matter: official text first, the open web only when you ask. Citations that do not resolve fail closed. Missing sources do not become a paraphrase.

    Unbound exploration is labelled separately from counsel work. Routing is described in settings as it actually runs — not as a residency slogan.

    #Compare and the corpus

    Jurisdiction compare and the primary-source corpus live in the same workspace. You can put treaties, competent authorities, and national rules side by side, then jump to the official text.

    A changes view lists what landed in the corpus recently. Inventory is not a verdict. COM(2025) 335 stays a proposal, not yet in force.

    #The open question stays put

    Folders keep threads with the matter. The question you are working stays visible while sources and drafts move underneath. Settings, profile, and activity sit in the same chrome — so the desk does not reset every time you switch a pane.

    Additional improvements (2)
    • Scholar-style browsing is available from Atlas when you need the teaching-depth corpus without leaving the matter.
    • Draft slots and the vault remain the place work product lives. Generated text is marked as AI-assisted.

    Learn more about Atlas

  2. Changelog

    You can now read how we classify our AI systems — and where inference is supposed to run.

    The legal pages publish our working EU AI Act classification, the Art. 50 disclosure, and the data-residency note. This is an internal classification with evidence, not a certification, not a CE mark, and not a regulator decision.

    #What the memo is

    The EU AI Act page states which Caelex surfaces are generative, what we tell users, and where the residual gaps are. Art. 50 copy lives next to it. Software is not legal advice.

    #What it is not

    It is not “AI Act certified”. It is not a promise that every inference path is EU-only in every configuration. Residuals are written down. Prefer the page over a slogan.

    Read the EU AI Act note

  3. Changelog

    You can put any two jurisdictions head to head — and turn a live orbit into a signed evidence pack.

    Scholar’s comparator places treaties, competent authorities, and national rules side by side and exports to PDF. Atlas adds full-depth research for more countries. Passage widens guided classification. Ephemeris turns a live orbit into a signed, offline-verifiable pack with probability bands — not false certainty.

    #Jurisdiction compare

    Pick two jurisdictions. Read the same questions across both. Export the view when you need it on paper.

    #Deeper Atlas coverage

    Australia, Japan, and Portugal are in the counsel corpus in full, with a deeper library of in-force ISO standards. Official links stay the source of truth.

    #Wider Passage classification

    Guided classification now covers propulsion, optics, propellant, power, launch, and re-entry. Incomplete data stays fail-closed. The officer stays in control.

    #Signed orbit evidence

    Ephemeris records a live orbit as a signed pack you can verify offline. Bands are probabilities. Validity limits are printed. It does not invent a clearance.

    Explore the platform

  4. Changelog

    Passage now covers the full export-control loop — and every consequential result explains itself.

    Classify, screen, license, ship. Four-eyes classification and a one-click “Why this?” dossier sit on the same path. AI proposes changes only. A named human reviews and applies.

    #Why this?

    Every result states what was decided, why, with what confidence, and which sources applied. That dossier is the thing you hand to the next reviewer — not a chat transcript.

    #Screening that can hold a shipment

    Shipments stay held while any sanctions match is unresolved. Each screening dataset shows a visible “current as of” date. Unresolved is not a pass.

    #Filings with a live preview

    Draft customs filings for German ATLAS, US AES, and Destination Control Statements with a live preview. Missing identifiers are flagged before you file.

    Additional improvements (1)
    • AI suggestions arrive as pending proposals. Nothing consequential applies itself.

    Learn more about Passage

  5. Changelog

    Passage now flags entities majority-owned by sanctioned parties — even when the name never appears on a list.

    Classification is sharper too: items match control-list parameters on their actual technical attributes. Screening always reflects the current official lists.

    #Ownership, not just the listed name

    Entities 50% or more owned by sanctioned parties are flagged automatically. A clean name is not a pass when the ownership chain is not.

    #Attributes the list actually uses

    Technical attributes are checked against control-list entries so the match is the parameter the list wrote — not a free-text guess.

    Get started

  6. Changelog

    Sanctions screening in Passage now spans multiple official lists in a single check.

    Items can carry structured technical attributes — the same parameters control lists are written in — so classification starts from precise data instead of free text. Every result stays checkable against the primary official lists.

    #One check, several lists

    A single screening run covers multiple official sanctions lists. The officer still sees which list hit, and how current it is.

    #Structured attributes on the item

    Record the parameters control-list entries actually specify. Classification starts there, not from a paragraph someone typed.

    Explore the platform

Entries describe customer-visible capabilities — not internal refactors, audits, or roadmaps. Software is not legal advice.

Changelog | Caelex