Submission 418
Mr Santhosh Kanaparthi
Submitter group: Individuals.
Position labels are not shown for individual submitters.
An AI model identified the submitter group and wrote the summary, key points and recommendation summaries. Counts do not measure importance or merit.
Summary
Read the original submission for the full text.
Proposes a statutory framework to limit AI-enabled behavioural profiling and exploitation of vulnerabilities, covering protected scope, permissible collection, approved purposes, separate consent, preserved access after refusal, verified deletion and reset rights, child protections, independent verification, criminal offences and whistleblower rewards. Presents these as reform proposals going beyond current privacy reform drafts.
Key points
Quotes are the submitter’s own words. Text taken from documents, including scanned pages read by text-recognition software, can differ from the PDF. Check the original. Listed in the same order as on the submission’s page in the interactive guide.
Argues that inferred vulnerability can be wrong yet affect treatment, so privacy should protect profiles and predictions as well as source data.
The searches are observations; vulnerability is an inferred conclusion. It could be wrong and still affect their treatment. Privacy therefore needs to protect profiles and predictions as well as their source data.
Argues that a profiling business model should not by itself establish a permissible collection purpose.
“Profiling is our business” must not, by itself, establish a permissible collection purpose. This safeguard should apply regardless of the organisation’s size, chosen technology or claimed risk level.
Argues that Parliament, not organisations, should define permitted purposes and their boundaries.
Parliament should define permitted purposes and their boundaries; organisations should not create their own exceptions. Any delegated approval power should operate within explicit statutory limits and independent oversight.
Argues that refusing optional consent should preserve registration, access and every feature that can reasonably operate without the refused processing.
Refusal should preserve registration, website or application access, and every feature that can reasonably operate without the refused processing. Only a feature genuinely dependent on that processing may be unavailable.
Argues that children should have age-appropriate rights and an additional review and reset opportunity at adulthood, supplementing lifelong deletion rights.
Children should have age-appropriate rights and assistance, plus an additional review and reset opportunity at adulthood. This should supplement lifelong deletion rights and prevent childhood commercial profiles becoming unchallengeable.
Argues that independent examination of systems and data flows should be required, and that self-declarations are insufficient.
The framework should require independent examination of systems and data flows before covered processing, after material changes and during operation. Organisations should maintain inventories, purpose and consent evidence, significant inferences, purpose-based access controls and tamper-evident logs. Self-declarations are insufficient.
Recommendations
Recommendations are shown beside the submission’s own words. Listed in the same order as on its page in the interactive guide.
Protect supplied, observed, linked and inferred information reasonably usable, alone or with available information, to single out, track or analyse someone’s patterns, circumstances or vulnerabilities, even without a name.
Protected personal data should include supplied, observed, linked and inferred information reasonably usable, alone or with other available information, to single out or track someone, or analyse their patterns, circumstances or vulnerabilities. A name need not be known: persistent device identifiers, accounts and behavioural profiles can enable targeting.
Cover deployed AI, machine-learning and rule-based systems capable of individual profiling or targeted content, offers, persuasion or decisions; include protected-data supply and connected workflows rather than exempting intermediate providers.
Covered processing should mean using that data in AI or other algorithmic systems that, in their deployed configuration, can profile individuals or enable individually targeted content, offers, persuasion or decisions. Relevant machine-learning and rule-based systems should be included, as should supplying protected data for that processing. Assessment should consider connected workflows, so one provider cannot escape coverage because another delivers the final message.
Do not trigger restrictions by the AI label alone; keep ordinary administration under collection/security/purpose rules, apply consent/deletion protections to personal-data training and approved-purpose requirements where training enables covered profiling.
The AI label alone should not trigger these restrictions. Ordinary collection, storage and administration remain subject to collection, security and purpose rules. Training on personal data should receive the separate consent and deletion protections below; training enabling covered profiling must also meet the approved-purpose requirement.
Protect transformed information remaining personal or reasonably linkable, despite synthetic/aggregate labels; assess linkability against held or reasonably accessible information and exclude genuinely non-identifying aggregates.
Protection should follow information that remains personal or reasonably linkable after transformation. Synthetic or aggregate labels should not defeat it. Re-identification and linkability should be assessed against information the organisation holds or can reasonably access. Genuinely non-identifying aggregate information should fall outside the protected scope.
Test collection purposes against objective statutory criteria before necessity; document purposes, demonstrate necessity, consider less intrusive alternatives, minimise collection and assess dataset combinations similarly.
Collection purposes should be tested against objective statutory criteria before necessity is considered. Organisations should document their purpose, demonstrate necessity, consider less intrusive alternatives and minimise collection. Combining datasets should face the same assessment.
Reject business-model justification alone regardless of size, technology or stated risk; reject accumulation for commercial demand, cheap storage, possible model improvement or unspecified future value, and collection without adequate protection.
“Profiling is our business” must not, by itself, establish a permissible collection purpose. This safeguard should apply regardless of the organisation’s size, chosen technology or claimed risk level. Commercial demand, cheap storage, possible future model improvement or unspecified future value should not justify accumulation. An organisation unable to protect the information adequately should not collect it.
Prohibit covered processing unless expressly lawful, demonstrably beneficial to the individual, necessary/proportionate in data use and compliant with consent/safeguards; have Parliament set boundaries and constrain delegated approval with independent oversight.
Covered processing should be prohibited unless it meets a purpose expressly permitted by law, delivers a specific and demonstrable benefit to the individual, uses necessary and proportionate data, and satisfies the consent and safeguard requirements. Parliament should define permitted purposes and their boundaries; organisations should not create their own exceptions. Any delegated approval power should operate within explicit statutory limits and independent oversight.
Do not justify processing by revenue or engagement alone or intrusive profiling by token user benefit; task permission must not authorise unrelated persuasion or profile sale.
A permitted service may also earn revenue for its provider. However, profit, advertising income or engagement alone must not justify processing, and a token user benefit must not excuse intrusive profiling. A user-requested spending analysis illustrates a potentially beneficial purpose; permission for that task would not authorise unrelated persuasion or sale of the resulting profile.
Allow Parliament to approve classes, including limited-intrusion routine personalisation, with clear conditions and demonstrable individual benefit; routine retail recommendations and segmented emails must remain protected.
Parliament may approve classes of activity, including routine personalisation with limited privacy intrusion, subject to clear conditions. This avoids separate applications for every company or feature. Retail recommendations and segmented emails using protected data must meet the applicable conditions, including demonstrable individual benefit; describing them as routine should not remove protection.
Statutorily prohibit exploitation despite approval or consent, including undermining choice through vulnerability and targeting children’s vulnerabilities; ban vulnerability-profile trading without proving purchaser intent, with narrow non-exploitative essential-care/protection exceptions.
Parliament should define exploitation that neither approval nor consent can authorise, including deliberate use of vulnerabilities to undermine meaningful choice for another party’s benefit and commercial targeting of children’s vulnerabilities. Trading vulnerability profiles should be prohibited without requiring proof of the purchaser’s intended exploitation. Any exception for essential care or protection must be expressly and narrowly defined in law and must not permit exploitation.
Apply approved-purpose requirements to all covered profiling; increase testing and oversight for higher risks while retaining basic protections for lower risks.
Approved-purpose requirements should apply to all covered profiling. Higher risk should strengthen testing and oversight, not remove basic protections from activities described as lower risk.
Separate legal permission from individual consent: require consent additionally where applicable and prevent signatures authorising prohibited conduct or replacing approved purposes.
Legal permission for a category of processing should not automatically authorise its use on every individual. Where required, consent should be an additional condition. Conversely, a signature should never authorise prohibited conduct or replace a required approved purpose.
Limit individual-benefit/consent exceptions to narrow express Parliamentary provisions requiring necessity, proportionality, limited duration and independent oversight; general public-interest, safety, research or government status is insufficient; waive consent only explicitly.
Any exception to individual benefit or consent should be expressly and narrowly defined by Parliament, with necessity, proportionality, limited duration and independent oversight. A general claim of public interest, safety, research or government status should not be sufficient. A purpose approval should waive consent only where the legislation explicitly says so.
Consider fraud prevention, public health, safety and ethics-approved research only under express statutory conditions; beneficial labels or ethics approval must not suffice or authorise unrelated vulnerability profiling.
Candidate purposes include fraud prevention, public health, safety and ethics-approved research. Each should qualify only within an express statutory permission and its conditions; a beneficial label or ethics approval alone should not suffice. For example, permission to detect fraudulent transactions should not authorise customer vulnerability profiling for unrelated sales.
Require separate consent for personal-data AI training, behavioural profiling, sensitive inference, sale, third-party disclosure and personalised persuasion, subject only to express exceptions; primary-business status must not exempt them.
Subject only to such an express exception, separate consent should be required for AI training on personal data, behavioural profiling, sensitive inference, sale, third-party disclosure and personalised persuasion. Calling one of these uses the organisation’s primary business should not remove that requirement.
For significant uses, require separate signed/electronically signed declarations, approximately one clear page, specifying information, purpose, processing, inferences, recipients, deletion date/event, refusal consequences and withdrawal.
For these significant uses, consent should take the form of a separate signed or electronically signed privacy declaration. Its essential explanation should fit on approximately one clear page and identify the information, purpose, processing, possible inferences, recipients, deletion date or event, consequences of refusal and withdrawal process.
Provide separate choices for materially different uses; reject silence, continued use, preselected boxes or general terms as consent; ensure accessibility, assess informed/free choice and avoid unnecessary identity documents or biometric signatures.
Materially different uses should have separate choices. Silence, continued use, preselected boxes or acceptance of general service terms should not substitute for the declaration. A signature is evidence of a choice, not proof that the choice was informed or freely made. Consent processes should be accessible and should not collect unnecessary identity documents or biometric signatures.
Make withdrawal equally straightforward and stop affected processing except under express legal exceptions; scrutinise employment/essential-service power imbalances and prevent data/business acquisitions automatically expanding or transferring permissions for new uses.
Withdrawal should be as straightforward as giving consent and should stop the affected processing, subject only to an applicable express legal exception. Power imbalances, including in employment and essential services, require particular scrutiny. Buying data or acquiring a business should not automatically expand or transfer permission for new uses.
Preserve registration, access and reasonably independent features after refusal; permit withholding only genuinely dependent features, with demonstrated rather than manufactured dependencies and no unrelated withholding or coercive degradation.
Refusal should preserve registration, website or application access, and every feature that can reasonably operate without the refused processing. Only a feature genuinely dependent on that processing may be unavailable. Organisations should demonstrate the dependency and must not manufacture it, withhold unrelated features or deliberately degrade service to pressure consent.
Retest purpose and necessity and obtain fresh consent where required for materially different uses; prevent automatic training, advertising or unrelated-sale reuse and outsourcing/sharing expanding permission.
Information collected for an approved task should not automatically become training material, advertising intelligence or an asset for unrelated sale. A materially different use should pass the purpose and necessity tests again and obtain fresh consent where required. Outsourcing or sharing should not expand permission.
Provide accessible data/profile access, purpose/inference/recipient information, correction and consequential-use challenge; human reviewers need information, time and outcome-changing authority; do not present uncertain predictions as established facts.
People should have accessible rights to know what data and profiles are held, understand their purposes, significant inferences and recipients, correct inaccuracies and challenge consequential use. Meaningful human review requires a reviewer with the information, time and authority to change the outcome. Uncertain predictions should not be presented as established facts.
Specify deletion dates/events before collection and automatically delete protected data and profiles when permitted purposes end.
Before collection, organisations should specify a deletion date or completion event. Protected data and profiles should be deleted automatically when the permitted purpose ends.
Provide lifelong reset covering histories, linking identifiers, inferences and scores; stop affected profiling, prevent historical-copy reconstruction and require fresh compliance for later collection/processing.
Separately, individuals should have a reset right at any age covering histories, linking identifiers, inferences and scores. A reset should stop affected profiling, prevent reconstruction from historical copies, and require fresh compliance for subsequent collection or processing.
Extend deletion to contractors, affiliates, brokers and overseas providers with recipient notices, statutory deadlines and verified completion; require reviewable refusals and specific legal retention obligations with restricted access/duration and no unrelated reuse.
Deletion should extend to contractors, affiliates, brokers and overseas providers, with recipient notification, statutory deadlines and verifiable completion. Refusals should state a reviewable legal basis. Retention exceptions should identify the specific legal obligation, restrict access and duration, and prohibit unrelated reuse.
Require enforceable backup expiry, quarantine and deletion reapplication on restoration; delete retained profiles as well as sources, and avoid unnecessary sensitive copies as compliance evidence.
Backups should expire within enforceable periods, remain quarantined from ordinary use and reapply deletions on restoration. Retaining profiles after deleting source files is insufficient. Compliance evidence should avoid unnecessary sensitive copies.
Plan verifiable model remedies before training; deleting training records is insufficient, and retraining or model retirement may be needed if other measures cannot satisfy deletion duties.
Deleting training records does not establish removal from a model. Organisations should plan for verifiable remedies before training; retained protected information may require retraining or model retirement if other measures cannot meet deletion obligations.
Apply stronger collection, retention and design standards, including addictive-engagement restrictions, to services likely accessed by children; neither parental nor child consent should authorise prohibited exploitation.
Services likely to be accessed by children should face stronger collection, retention and design standards, including restrictions on addictive engagement. Neither parental nor child consent should legitimise prohibited exploitation.
Provide children age-appropriate rights and assistance and an additional adulthood review/reset, supplementing lifelong deletion rights and preventing unchallengeable childhood commercial profiles.
Children should have age-appropriate rights and assistance, plus an additional review and reset opportunity at adulthood. This should supplement lifelong deletion rights and prevent childhood commercial profiles becoming unchallengeable.
Base vulnerability protections on circumstances and conduct rather than demographic assumptions; retain permitted beneficial assistance while prohibiting deliberate exploitation of vulnerability, including dependency optimisation undermining choice, and assessing downstream exploitation risks.
Protections should follow the circumstances and conduct rather than assume that everyone in a broad demographic group is vulnerable. Beneficial assistance for a permitted purpose should remain possible. Deliberately using vulnerability to undermine meaningful choice for another party’s benefit should be prohibited even if collection was lawful. This prohibition should extend to deliberately using behavioural profiles or inferred vulnerabilities to optimise compulsive use or dependency in a way that undermines meaningful choice. Providers should assess foreseeable exploitation risks, including risks introduced through downstream access to profiles.
Require independent system/data-flow examination before covered processing, after material changes and during operation; maintain inventories, purpose/consent/inference evidence, purpose-based access controls and tamper-evident logs; escalate scrutiny under published risk criteria/assessor-independence rules.
The framework should require independent examination of systems and data flows before covered processing, after material changes and during operation. Organisations should maintain inventories, purpose and consent evidence, significant inferences, purpose-based access controls and tamper-evident logs. Self-declarations are insufficient. Scrutiny should increase with sensitivity, scale, persuasive capability and consequences under published criteria and assessor independence rules.
Resource regulator inspection, testing and unannounced checks; legislate a lead regulator, sector coordination, complaint, review and transition arrangements.
Regulators should be resourced to inspect, test and conduct unannounced checks. Legislation should identify a lead regulator, coordinate sector duties and establish complaint, review and transition arrangements.
Apply protections to people in Australia regardless of processing/incorporation location; require accountable Australian representatives for offshore providers, enforceable transfer-recipient duties and regulator evidence access; prohibit transfers where safeguards cannot be secured.
The framework should apply to protected data about people in Australia regardless of processing location or incorporation. Offshore providers should appoint an accountable Australian representative for notices, complaints and cooperation; transfers should require enforceable recipient duties and regulator evidence access. Transfers should not proceed where safeguards cannot be secured.
Have regulators use AI Safety Institute technical assessments to review foreign-interference, cyber-security and essential-service exposures as capabilities develop.
Concentrated offshore control of behavioural profiles could facilitate foreign interference; compromised data and dependence on overseas AI services could also weaken cyber security and essential-service resilience. These foreseeable risks connect the safeguards to ToR (l) and (m). As capabilities develop, regulators should use the AI Safety Institute’s technical assessments to review these exposures.
Retain deployer accountability despite outsourcing and no blanket government exemption; preserve corporate/individual responsibility, tested human intervention, suspension/shutdown for high-risk systems and essential-service fallbacks.
Outsourcing should not remove deployer accountability; government agencies should receive no blanket exemption. The framework should retain applicable corporate and individual responsibility, tested human intervention, suspension and shutdown for high-risk systems, and essential-service fallbacks.
Have Parliament define offences, fault, penalties and corporate/individual responsibility for prohibited data exploitation and vulnerability-profile trading, with knowing or reckless fault.
Parliament should define offences, fault requirements, penalties and applicable corporate and individual responsibility. Illustrative offence categories are: • Knowingly or recklessly supplying or processing protected data for exploitation expressly prohibited by statute, or trading vulnerability profiles contrary to the statutory prohibition.
Criminalise knowing/reckless covered processing without required permission/consent or materially outside authorised purposes/conditions, causing or seriously risking significant harm or involving systematic conduct at scale, subject to express statutory exceptions.
Knowingly or recklessly conducting covered processing without required permission or consent, or materially outside authorised purposes or conditions, where it causes or creates a serious risk of significant harm, or involves systematic conduct at scale; subject to express statutory exceptions.
Criminalise protected-data supply for covered profiling or individual persuasion where the supplier knows or is reckless that the intended covered use lacks required statutory permission or consent.
Supplying protected data for covered profiling or individually targeted persuasion where the supplier knows, or is reckless as to whether, the intended covered use lacks a required statutory permission or consent.
Criminalise deliberate concealment of prohibited data flows, falsified required evidence, obstructed lawful inspection or defeated controls enabling prohibited processing.
Deliberately concealing prohibited data flows, falsifying required evidence, obstructing lawful inspection or defeating controls to enable prohibited processing.
Cover internal access, connected databases, training and external providers involved in prohibited conduct; base individual liability on conduct/fault rather than title, protect authorised testing/responsible disclosure and distinguish deliberate destruction from lawful deletion.
The framework should cover internal access, connected databases, training and external providers involved in prohibited conduct. Individual liability should depend on conduct and fault, not job title. It should protect authorised testing and responsible disclosures and distinguish deliberate evidence destruction from lawful scheduled deletion.
Define significant harm, scale and fault for each offence; retain proportionate civil penalties, compensation, correction, deletion and suspension for smaller breaches; transfer data duties with assets without automatically criminalising innocent purchasers.
Parliament should define significant harm, scale and fault for each offence element. Smaller breaches remain subject to proportionate civil penalties, compensation, correction, deletion and suspension orders. Data obligations should follow transferred assets without automatically criminalising an innocent purchaser.
Confine supply offences to knowing/reckless facilitation of specifically unauthorised covered uses, without additionally requiring significant harm or systematic conduct at scale.
The supply offence is confined to knowing or reckless facilitation of a specifically unauthorised covered use and therefore does not additionally require proof of significant harm or systematic conduct at scale.
Provide employees, contractors and researchers confidential channels, anti-retaliation protection and practical assistance; protect honest concerns even without proven breaches and avoid deterring responsible reports through fabricated-evidence safeguards.
Employees, contractors and researchers should have confidential reporting channels, protection against retaliation and practical assistance. Honest concerns should remain protected even if an investigation finds no breach; safeguards against knowingly fabricated evidence should not deter responsible reporting.
Reward eligible disclosures materially contributing to successful enforcement with an independently administered statutory percentage of recovered fines/penalties; Parliament sets eligibility, percentage, allocation and review rights; exclude company payment control and integrate existing protections.
Eligible disclosures that materially contribute to successful enforcement should attract an independently administered, legally defined percentage of fines or monetary penalties actually recovered. Parliament should set eligibility, the percentage, allocation between informants and review rights. Companies should not control payment, and the scheme should integrate with existing protections.