A

Amplified

A search starts with a plain language description of an invention or a patent number, and the system sorts more than 140 million records against it in real time by conceptual similarity, in one of two published AI ranking modes. Relevance feedback is the core loop: results marked relevant are fed back so the system surfaces more like them, and machine learning can partly automate relevance review of new publications.

Boolean search, proximity operators and field selection are supported as ways to steer the ranking, not as what is being bought; Amplified says its AI reads the patents so the user does not have to. Without the model, what remains is a patent database with a query box.

The product returns records from a defined collection rather than generating text, so there is little room for fabrication in the usual sense. A published design commitment goes to defensibility: the AI sorts results but never excludes them, so a ranking error cannot silently remove art from a searcher's view. Keyword counts are shown so a user can spot where the model misread an input, and search history can be exported.

The agreement acknowledges failure: section 19 says outputs may be inaccurate or incomplete, and section 14 says patent legal status is shown as received from authorities without warranty. No current, testable figure is published. Amplified's blog records a 2019 EPOPIC study reporting automated prior art results comparable to professional search across nearly 1.5 million patent search reports. That study is seven years old and has not been repeated for the current product. Clause 4.2(xxi) prohibits benchmarking or comparing the Service for publication without prior written consent.

The AI sorts but never excludes, so the person decides what to stop reading rather than the model deciding what they never see. The working loop in the help center is driven by the searcher, who marks results relevant, flagged or hidden, iterates, and runs a final check with filters removed to catch anything missed. Keyword counts let the user spot the model's misunderstandings. The agreement requires supervision: clause 4.2(xix) prohibits using outputs in or as the basis for legal, medical, financial, safety critical or other regulated decisions without independent review by a qualified professional, and 4.2(xx) prohibits relying on outputs in place of the customer's own due diligence.

No confidence signal is published, no threshold at which the system defers is stated, and nothing covers what happens when a search is later shown to have missed art.

Amplified's home page carries a quotation praising the product with no name, role or organization. The blog records conference activity rather than deployments: an EPO keynote at Search Matters 2019, an EPOPIC 2019 presentation, an invitation to speak at LESI, and selection among five of ninety startups at NIKKEI's AI SUM. Those show standing in the field, not customer outcomes. Third party sources, including LOT Network, CB Insights, WIPO and a revenue database, report customers such as Ricoh, Unicharm, Mizkan Holdings and a European automaker, and more than fifty large enterprises in thirteen countries.

None of that comes from the vendor, and no case study, named customer or dated deployment figure is published on any vendor page.

Clause 6.3 says Amplified will never use customer content to train general AI models, and allows customer specific models or classifiers only at the customer's direction and only to provide those features back to that customer. The content license is limited to its purpose and to the period of use, ending when content is deleted from active systems. Retention and deletion are specific under 6.5, 12.9 and 16: deletion from active systems within thirty days of termination or request, written confirmation on request, and thirty days to export.

A mutual confidentiality clause in section 13 covers information reasonably understood to be confidential. Access control is contractual under 12.4, with personal credentials and an account administrator. Nothing addresses privilege or attorney work product, although projects hold invention disclosures and internal ideas. No model provider is named, so what any underlying provider may keep is not stated.

Section 14 of the terms, headed No Professional Advice, says professional information through the Service is for information only and not advice. It then names the concrete failure case for a patent tool: patent legal status is shown as received from authorities without warranty, and users are directed to the national patent authority for authoritative status. Supervision is handled in the acceptable use policy: clause 4.2(xix) prohibits using outputs in or as the basis for regulated legal decisions without independent review by a qualified professional.

Use is limited to adults acting in a business or professional capacity, not consumers. No jurisdiction limit is stated for the product's own coverage, and the direction to a national authority addresses legal status only.

No governance position is published for a system that decides which prior art a professional sees. There is no responsible AI statement, governance framework, named owner for model behavior, evaluation regime for ranking quality, published test results, or anything on bias. Bias has a specific shape here: a similarity model offering native AI search only in English and Japanese may perform unevenly across the 22 languages and 97 authorities it covers, and nothing addresses that.

The information security program aligned with ISO 27001 is a security matter, not AI governance. One clause works against outside scrutiny: clause 4.2(xxi) prohibits benchmarking, testing or comparing the Service for publication without prior written consent.

On retention and deletion, clause 12.9 commits to deletion from active systems within thirty days of termination or request, with written confirmation available; 6.5 adds deletion from backups in the regular cycle; and section 16 gives thirty days to export before deletion. On access control, 12.4 requires access controls, encryption in transit and at rest, logging and monitoring, vulnerability management and secure development, with confidentiality duties on authorized staff, and section 3 makes credentials personal under a named account administrator.

On incidents, 12.8 commits to notice without undue delay, covering the nature of the incident, categories and approximate volume affected, likely consequences and measures taken. On subprocessors, 12.5 commits to publishing a current list on the trust center, with email updates of changes and equal protections passed down. Transfers are covered at 12.6, audits at 12.10 and US state privacy terms at 12.12.

Section 18 provides the Service as is and as available, with all warranties disclaimed, including fitness and noninfringement. Section 19 caps total liability at fees paid in the twelve months before the event and excludes indirect and consequential damages. It also removes the product risk rather than allocating it: the customer accepts that outputs may be inaccurate, incomplete or inappropriate, is solely responsible for checking them, and Amplified disclaims all liability for reliance on outputs.

The indemnity in section 17 runs only from customer to Amplified, with no intellectual property or other indemnity to the customer. No insurance position is published. Carve outs are kept for death, personal injury, fraud and willful misconduct.

The help center documents importing a result list from a complementary search tool, naming CAS and STN, a specific path for a professional searcher. The site navigation carries How it works, Blog, Pricing and Resources, with no integrations page, API reference or developer documentation. A public GitHub organization is linked from the footer. No IP management system, docketing platform, document management system or patent office filing interface is named on any vendor page.

Single sign on providers appear only in a third party software directory. Export is well served, with search history export and sharing to unlimited viewers, but export is not integration.

Amplified's trust center says it builds on Fly.io and Amazon Web Services and relies on their compliance with privacy and security standards, including regular independent audits, with data encrypted in transit and at rest. Naming providers is not naming a location. No region, country or data center appears anywhere, no residency option is offered, and nothing says whether the platform is single or multi tenant or whether enterprise customers get a dedicated arrangement.

Clause 12.6 commits to a lawful mechanism for international transfers of personal data, naming EU Standard Contractual Clauses Module Two, the UK Addendum and IDTA, a Swiss addendum and Brazilian SCCs. That shows data crosses borders without saying between which places.

Amplified claims no certification of its own. Clause 12.4 describes an information security program aligned with ISO 27001 standards, and the trust center describes adherence to ISO compliance standards. Alignment is self assessed: no certifying body, scope, observation period or report date is published, so nothing can be checked against a register. The certifications named belong to the hosts: the trust center cites the compliance of Fly.io and Amazon Web Services and their regular independent audits, which does not carry over to the platform.

The route to evidence is a trust center at trust.amplified.ai, a commitment in clause 12.4 to publish security summaries and independent assessment information there, and a commitment in 12.10 to provide audit summaries and security documentation on written request.

The training clause separates general AI models from customer specific models and classifiers that Amplified trains, and the product has two AI ranking modes, semantic and neural concept similarity, built on deep learning over patent data. So Amplified runs its own models rather than reselling a general assistant. Nothing verifiable is published about them: no model, version or architecture, no third party model or inference provider, no location for inference, and no commitment to notify customers of model changes.

Fly.io and Amazon Web Services are named on the trust center as infrastructure, which says where the platform runs, not whose models process a search. Clause 12.5 commits to publishing the subprocessor list on the trust center.

The rate is from $500 a month paid annually, a floor rather than a flat rate, and what it buys is listed: unlimited projects and viewers, search across more than 140 million patents, sharing of up to 50 projects, discussion and annotation, and cancellation at any time. The intended uses are named as novelty, invalidity and freedom to operate, and both AI ranking modes are included. Purchase is self serve, with signup and a trial request published and no sales conversation needed.

Team pricing and enterprise options are quoted, not published, and clause 9.1 points to a second pricing page inside the application. The agreement adds that subscriptions renew automatically unless canceled in account settings, payments are not refundable except where law requires, and fees may change going forward on notice.

Searchable scope is quantified: more than 140 million records, 97 or more authorities, 22 languages, about 34 million original language publications and 50 million machine translations in the English index, plus family data and citations both ways. The work is named as novelty, invalidity and freedom to operate, the practice's own vocabulary. Buyers named include IP law firms, corporate research and development and IP teams, patent offices, universities and individual inventors.

One limit is stated plainly: AI search is native in English and Japanese only, with other languages in beta and on request. No firm size or seat threshold is addressed, and in house and government use are named without description.