Product and intended use
Define the configuration, people, information and tasks being assessed.
Release stage
Public claim only
The disclaimers page (last updated 5 March 2025) states IP First Response 'is a new pilot service being trialled by IP Australia. It is still in development.' The 4 May 2026 news item says IP Australia 'is piloting' the chatbot and that it 'is currently in its early stages of rollout'. The site's own logo files are named alpha and beta; no page read states which stage the chatbot is at now.
Ask: Has the chatbot moved past pilot or beta, and what is the published criterion and date for general availability?
What the chatbot does
Public claim only
The 'About the AI Chatbot' page (22 April 2026) describes retrieval-augmented generation: the question is first assessed for whether it is in remit (the chatbot may say it cannot help or ask follow-up questions), then rewritten in several forms and turned into vector embeddings, then matched by meaning rather than keywords against the corpus, then an AI model reads the retrieved material and writes the response, links the exact pages used and pre-writes follow-up questions for selection. A feedback option sits at the bottom of each response. The platform's Navigator tool is described as a filter-based alternative 'with no AI involved'.
Ask: Which parts of that pipeline are configurable, and is the remit assessment step documented anywhere a user can inspect?
Jurisdiction and corpus
Public claim only
The operator states the chatbot returns 'Australia specific information' and draws only on IP First Response content about IP enforcement options in Australia plus a curated selection of IP Australia public web pages. It is not described as covering case law, legislation as enacted, or any jurisdiction other than Australia. Currency of the corpus is not stated.
Ask: How often is the corpus refreshed, and how is a user told the date to which the material is current?
Vendor content addressed to AI systems
Public claim only
Checked on 2026-09-27 across the chatbot, developer, privacy, disclaimer and news pages and a URL map of the site: no page or block was found telling AI models, assistants or crawlers how to describe the product, and no /llms.txt, /ai-info or /for-ai URL appeared in the map. The site does publish a page addressed to AI builders at /developers, headed in part 'Use IP First Response content in your own AI work', which names 'People exploring chatbot tools (ChatGPT, Copilot, Claude, etc.)' among its audience, offers a feed at /options-data-feed 'designed to plug into AI tools', and states: 'Where content from this website is used for model training, please attribute the IP First Response website.' That page invites reuse of content; it does not direct how the product is to be characterised.
Ask: Is the /options-data-feed feed versioned and dated, and does IP Australia give any currency or accuracy commitment to a firm that embeds it in its own assistant?
Data, controls and requirements
Compare the actual configuration with firm, client and jurisdiction requirements, supported by evidence.
What may be typed in, and on whose authority
Public claim only
The Chatbot Privacy Notice (22 April 2026) instructs users not to enter personal information (names, contact details, business identifiers, information about an identifiable individual, sensitive information), commercially sensitive information, or legally privileged or confidential material, and not to ask the chatbot to generate personal information. It states that anything entered in the free-text fields will be collected and handled by IP Australia, that by entering information the user consents to that handling, and that the user warrants they have 'the appropriate authorisation to permit such handling'. Selecting 'Start chat' is the consent gate, and access is refused without it.
Ask: For a practitioner using the chatbot on a live client matter, does IP Australia accept matter facts that could identify a client, and what happens to a conversation in which privileged material is entered by mistake?
Foundation model and provider
Public claim only
The 'About the AI Chatbot' page (22 April 2026) states: 'We use an AI model from Anthropic, hosted securely in Australia with Amazon Bedrock, to read the retrieved information and write a response that answers your query while adhering to guidelines we've set.' The model name and version are not published, no embedding model is named, and no page read describes any user or organisational model choice. The privacy notice refers generally to 'third-party technology providers' without naming them.
Ask: Which Anthropic model and version is in production, which model produces the embeddings, and what change notice is given when either is replaced?
Third-party providers
Public claim only
No subprocessor list is published. The chatbot page names Anthropic as the model provider and Amazon Bedrock as the hosting service. The Chatbot Privacy Notice states that where third-party technology providers are involved IP Australia may disclose personal information to them for the purpose of providing the service, and applies 'contractual, technical and governance controls'. The platform privacy collection statement (25 March 2025) names Mixpanel for anonymous website and tool analytics using first-party cookies. Feedback forms are hosted on a Qualtrics domain.
Ask: Provide the current list of third-party providers for the chatbot, with entity, country, purpose and the data each receives, including analytics and feedback-form providers.
Training on user content
Public claim only
The Chatbot Privacy Notice states that, subject to disclosures required by law, personal information entered into the chatbot is not used 'to train or fine-tune AI models', and separately that 'User inputs are not used to train AI models.' The same notice permits limited records to be retained for quality improvement, improving IP Australia content and services, and informing service design and policy development. The notice is the consent document a user accepts to use the chatbot; there is no separate customer contract.
Ask: Is the no-training position also binding on Anthropic and Amazon Web Services by contract, and does it cover fine-tuning, evaluation datasets and prompt-improvement sets as well as model training?
Feedback and quality monitoring
Public claim only
The chatbot page states there is a feedback option at the bottom of each response and that 'It directly informs how we improve the system.' The privacy notice lists service assurance and monitoring, quality improvement, troubleshooting and service design among the purposes for which limited records may be retained, and states that where practicable personal information may be de-identified. It does not say whether a flagged response is retained with its conversation, or for how long.
Ask: When a user submits negative feedback on a response, what is stored with it, who reviews it, and how long is it kept?
Retention of conversations
Public claim only
The chatbot page states the chatbot is stateless and that nothing from a conversation is retained in the browser after the session ends, and that before storage the conversation is run through an algorithmic filter that removes 'the most common types of personally identifiable information like IP registration numbers and email addresses'. The privacy notice states automated controls remove personal information from long-term storage 'where possible', that the process 'may not be 100% effective', and that limited records may be retained for named purposes. No retention period is published in either document.
Ask: What is the retention period for stored chatbot interaction records, and what is retained where the automated filter fails to remove personal information?
Deletion of stored records
Not confirmed
Neither the chatbot page nor the Chatbot Privacy Notice describes a way to have a stored conversation record deleted, and there is no account to delete because the chatbot is used anonymously. The privacy notice points to IP Australia's agency Privacy Policy for access, correction and complaints; that policy was not reached within the operation budget and is at https://www.ipaustralia.gov.au/about-us/our-agency/privacy.
Ask: How does a user ask for a stored chatbot interaction record to be deleted, given that conversations are anonymous and cannot be identified by an account, and what is the timeframe?
Who can read stored conversations
Public claim only
The chatbot page states interaction data 'is held in Australia, protected by access controls, audit logging, and encryption' and that IP Australia does not sell or share the data. The privacy notice states records may be retained for troubleshooting and service assurance and that personal information may be disclosed to third-party technology providers for the purpose of providing the service. Neither document states which roles inside IP Australia may read stored conversations, or whether provider staff can.
Ask: Which roles inside IP Australia, and which provider personnel, can read stored chatbot conversations, and is that access logged and reviewable?
Access control and administration
Not confirmed
The chatbot is a public endpoint used without an account: the operator states 'You don't need to create an account or tell us who you are.' There is therefore no authentication, no role model and no administrative console. The IP Australia security statement describes username and password verification for its own logged-in online services, which are separate from the chatbot. Nothing in the pages read lets an organisation restrict, log or administer staff use.
Ask: Is there any way for an organisation to see, restrict or log its staff's use of the chatbot, or is it purely an unauthenticated public endpoint?
Encryption
Public claim only
The chatbot page states interaction data held in Australia is protected by encryption, without naming algorithms or distinguishing transit from rest. The IP Australia security statement says the agency follows the Australian Government Cyber Security Standards and encrypts information sent between the user's computer and its servers using SSL/TLS when logging in to its online services. No key-management statement was found for the chatbot.
Ask: Name the encryption in transit and at rest for chatbot conversations and the key-management arrangement, including for data passed to the model provider.
Independent security assurance
Not confirmed
No independent assurance report is published for the chatbot. The IP Australia security statement (read 2026-09-27) says the agency follows the Australian Government Cyber Security Standards and describes encryption and password protection; it names no SOC 2 report, ISO/IEC 27001 certificate or IRAP assessment. The site publishes a vulnerability disclosure policy, linked from that statement.
Ask: Has the chatbot undergone an IRAP assessment or other independent security assessment, and can the report or an assessment summary be provided?
Terms, liability and forum
Public claim only
There is no customer contract: use is governed by the disclaimers page (5 March 2025) and the consent gate in the Chatbot Privacy Notice. The disclaimers state IP Australia 'makes no warranty or representation that the material is complete or free from error' and accepts 'no legal responsibility regarding the accuracy and reliability of material on this website', and no liability for interference with or damage to the user's computer, software or data. No service level, indemnity, governing-law clause or liability cap appears. The privacy notice is expressly subject to the Privacy Act 1988 (Cth) and the Australian Privacy Principles.
Ask: Is any service commitment, indemnity or written assurance available to an organisation relying on the chatbot, or is use entirely at the user's risk under the disclaimer?
Cross-border processing
Public claim only
The Chatbot Privacy Notice states: 'It is possible that personal information may be temporarily held overseas when being processed, including as facilitated by relevant third-party technology providers. However, such personal information will only be stored on a persistent basis in Australia.' The chatbot page separately states the model is 'hosted securely in Australia with Amazon Bedrock' and that interaction data is held in Australia. The two statements are not necessarily inconsistent, but the notice leaves open which processing may occur overseas and in which country, and no country is named.
Ask: Which processing steps may occur outside Australia, in which countries, and does model inference itself run only in an Australian region?
Incident notification
Not confirmed
No breach-notification commitment, timeframe or incident history was found in the Chatbot Privacy Notice, the platform privacy collection statement, the disclaimers page or the IP Australia security statement. The privacy notice states that personal information is protected by law including the Privacy Act 1988 (Cth) and the Australian Privacy Principles, and that IP Australia may disclose personal information where required or authorised by a court or tribunal order, under an Australian law, or compelled by Parliament.
Ask: How will users be told if stored chatbot interaction records are involved in a data breach, and within what period from discovery?
Change notice
Not confirmed
Each published page carries a 'Last updated' date (chatbot page and privacy notice 22 April 2026, developers page 22 May 2026, disclaimers 5 March 2025, platform privacy statement 25 March 2025), but no page read describes a change-notification mechanism, version history or notice period for changes to the model, the corpus, the privacy notice or the disclaimers. The chatbot page says the operator will 'keep iterating based on what we learn'.
Ask: How will a material change to the model, the corpus or the privacy notice be notified, and is a version history of the privacy notice available?
Export and exit
Public claim only
The chatbot page states: 'The chatbot is stateless. Once your session ends, nothing from that conversation is retained in your web browser. If you want to keep your conversation, make sure to download it!' The download format is not stated. There is no account, so there is no account export; no organisational export of stored interaction records is described.
Ask: In what format does a conversation download, and can an organisation obtain or delete the interaction records associated with its use?
Connectors and content feed
Public claim only
The chatbot itself is not described as connecting to any external system. In the other direction, the For Developers page (22 May 2026) publishes the whole platform content as a structured feed at https://ipfirstresponse.ipaustralia.gov.au/options-data-feed, described as 'a structured version of all content, designed to plug into AI tools', and names internal IP assistants, user-facing chatbots and in-product guidance as things it could be used to build. The feed itself was not fetched.
Ask: Is the options data feed stable, versioned and dated, and what notice is given before its structure changes?
Open-web access
Public claim only
The chatbot page states: 'The chatbot doesn't have access to the open internet. If the answer isn't on our shelf, it should tell you it doesn't know.' The second sentence is conditional as published. The same page states that a question outside the tool's remit may prompt the chatbot to say it cannot help or to ask follow-up questions.
Ask: Is the refusal behaviour enforced by the system or dependent on the model following instructions, and has the rate of out-of-corpus answering been measured?
Agent actions
Public claim only
Nothing in the chatbot page, developers page, privacy notice, disclaimers or news item describes the chatbot taking any action beyond returning text, links to the pages it used, and pre-written follow-up questions for the user to select. No email, filing, lodgement or write-back to any IP Australia system is described, and the chatbot is separate from the agency's transactional online services.
Ask: Does the chatbot, now or on the roadmap, take any action beyond returning text, for example lodging, pre-filling or emailing on a user's behalf?
Human review before reliance
Public claim only
The Chatbot Privacy Notice states that the chatbot 'may get some information incorrect or otherwise present information in a false or misleading way', that a user seeking information important to their situation 'should always check the original source', and that 'You are responsible for validating any information presented to you by the chatbot before relying on it.' The chatbot links the exact pages used so a user can check them. No operator-side review of individual responses before display is described.
Ask: What monitoring or sampling of chatbot responses does IP Australia perform after release, and are corrections published when a response is found to be wrong?
