Block8.ai utilise Anthropic's latest suite of Claude models, including Opus, Sonnet, and Haiku, accessed via AWS Bedrock through a Private Link.

Block8.ai uses Anthropic's Claude models through AWS Bedrock via a Private Link to our VPC. We maintain a zero data retention policy for all evidence collected during the testing process. AWS Bedrock is in scope for common compliance standards including ISO, SOC, CSA STAR Level 2, is HIPAA eligible, and can be used in compliance with the GDPR.

Block8.ai ensures deterministic outcomes and consistent reporting of vulnerabilities and exploits through:

• A Chain of Experts/Agents framework, where findings are audited at each stage to minimise false positives.

• An extensive library of deterministic code for common actions, known vulnerabilities, and exploits to ensure repeatable testing.

• Integration with open-source tools for scanning and exploitation, producing structured outputs.

• MCP servers capturing all agent data and findings in a structured format.

• Supervision by qualified penetration testers who approve and guide agent outputs.

Block8.ai transmit only the minimum necessary data for decision-making and vulnerability analysis is sent. Data never leaves Block8.ai’s controlled network, is never transmitted over the public internet, and is encrypted both in transit and at rest. Block8.ai does not use this data to train AI models, does not maintain proprietary models, and does not perform prompt caching or storage.

Block8.ai processing follows ISO 27001, ISO 27017, and SOC 2 controls. Block8.ai is ISO27001 certified and in the observation period for SOC2 certification. The platform and its underlying services, including AWS Bedrock, are SOC2 and ISO27001 compliant, HIPAA eligible, and GDPR-compatible.

Yes. Block8.ai agents run in fully isolated container instances with a sandboxed environment. Instances are destroyed upon test completion along with any forensic data not required for reporting. The sandbox enforces a strict list of permitted and denied actions and has multiple guardrails at each stage.

All Block8.ai penetration testers are CREST or OSCP certified with at least 5 years of offensive security experience. Human supervision is mandatory: testers can halt agents at any time, and all stage transitions and tool executions require tester approval.

The Block8.ai Platform will require access to all areas that the client require testing.

• For an external penetration test, Block8.ai will attempt to gain access to as much of the client network as possible however initial access will only be to the external ports of internet facing technologies.

• For a web application penetration test, Block8.ai will require access to the front end of the application. They will then attempt to gain access to as much of the application as possible to achieve the necessary reporting.

No, Block8.ai is a solely Australian sovereign operation.

The client will only provide access and authorisation for the time that testing is to be conducted.

Block8.ai will notify the client immediately should Block8.ai become aware of any issue which poses a material threat to the client.

Yes, Block8.ai does not store customer data identified during testing and would only retain such information that is necessary for reporting the current engagement or re-tests/future engagements. Block8.ai supports secure deletion wherever the deletion of data is necessary.  For the majority of instances, Block8.ai utilises the AWS secure deletion utilities provided with the platform.

Block8.ai utilise fine-tuned models and Anthropic's Claude models through AWS Bedrock via a Private Link. We also maintain zero data retention policy with the services. Bedrock is in scope for common compliance standards including ISO, SOC, CSA STAR Level 2, is HIPAA eligible and can be used in compliance with the GDPR.

Anonymised metadata from exploitation may be used to train future models or fine tune isolated environments for performance improvements and cost reductions.  The client can choose to opt out at any time.

The services utilised by Block8.ai fall under AWS Data Protection policy (https://docs.aws.amazon.com/bedrock/latest/userguide/data-protection.html) assures Amazon Bedrock doesn't store or log your prompts and completions. Amazon Bedrock doesn't use your prompts and completions to train any AWS models and doesn't distribute them to third parties. All data is processed on AWS infrastructure via private network links and is always encrypted in transit and when at rest. Isolated instances are available on request at an additional cost.

Prompts and commands are only able to be input internally within Block8.ai, this is not available to persons outside of the organisation and access is not permitted. Additionally all client data and testing processes are fully sandboxed with separate database instances running in isolated containerised environments and networks so LLM agents do not have awareness or access outside of their sandbox.

Block8.ai humans are currently able to have oversight across all phases of testing, we employ a number of professional and highly qualified penetration testers to oversee the process. As an additional measure at regular intervals, Block8.ai conducts parallel testing using AI and human testers as a means of assuring the quality and comprehensiveness of the testing process.

The Block8.ai Platform is built using the following design principles:

• Containerised environments per client hosted on separate servers and subnets with encrypted drives. This keeps client data segregated and secure and prevents any potential data leaks.

• Separation of Concern principles used for services. This means services and AI Agents only have access to what they need and are never overprivileged.

Block8.ai applies the following practices to prevent AI going rouge on your systems:

• Block8.ai’s Execution Engine allows Human in the Loop intervention for potentially dangerous actions defined by a curated whitelist of commands. It will not execute commands that have not been previously whitelisted and halt before requesting human review and approval.

• AI Prompts are seeded with curated Guardrails and created with context engineering best practices applied to ensure:

  • context window overflow does not occur,

  • important details are never compacted,

  • context state is correctly managed.

• AI Agents do not have administrative access to the underlying Operating Systems.

Block8.ai have implemented a Triple Human Validation. A senior Cyber SME (Penetration Testers and/or GRC) will validate the following AI driven outcomes:

• Vulnerability Scanning Results

• Vulnerability Exploitation

• Vulnerability Recommendation (aka Reports)

Whilst Block8.ai does not store client data the Block8.ai system is built using the following design principles:

• Containerised environments per client hosted on separate servers and subnets with encrypted drives. This keeps client data segregated and secure and prevents any potential data leaks.

• Separation of Concern principles used for services. This means services and AI Agents only have access to what they need and are never overprivileged.

• Block8.ai utilise AWS provided encryption facilities across all systems.

Block8.ai has a suite of policy documents that have been approved by the board and communicated to all employees. This suite of policies was examined by and approved by the CREST organisation as part of Block8.ai's certification process. Additionally, governance documentation is being expanded to include all requirements in Block8.ai's preparation for ISO27001 certification.

Block8.ai services offer unrivalled privacy levels fully compatible with national privacy requirements, which are hosted and run entirely within Australian national boundaries.

The Block8.ai board monitor all regulatory compliance in accordance with the Board Charter. Security programs can be adjusted as required once approved and documented by the board.

Whilst not mandated, penetration testing is strongly recommended for compliance with the Privacy Act in Australia. Block8.ai T&Cs maintains your privacy and confidentiality. Block8.ai does not store or collect personal information other than client contact details.

All compliance standards require that you better understand your environment, that you identify vulnerabilities and remediate them in a timely manner. Penetration tests are generally required or recommended for compliance with:

• Payment Card Industry Data Security Standard (PCI DSS)

• ISO 27001

• Essential Eight framework

• Security of Critical Infrastructure Act (SOCI)

• Privacy Act 1988 (and subsequent amendments)

• SOC 2

• EU General Data Protection Regulation (GDPR)

• US Health Insurance Portability and Accountability Act (HIPAA)

Block8.ai has aligned its risk assessment framework to ISO31000 and ISO27005.