I beg to move Amendment 1, in Clause 15, page 22, line 15, at end insert—

“(f) whether the incident involves failure modes not previously observed in the relevant sector materially involving autonomous or adaptive systems based on machine learning, including where the potential impact of such failure modes was mitigated or prevented.”

With this it will be convenient to discuss the following:

Amendment 2, in Clause 15, page 22, line 25, at end insert—

“(ea) where the incident was associated with one or more autonomous or adaptive systems based on machine learning, details of those systems and their involvement in the incident;”

Amendment 4, in clause 15, page 23, line 32, at end insert—

“(ea) where the incident involved one or more autonomous or adaptive systems based on machine learning, details of those systems and their involvement in the incident;”

Amendment 5, in clause 15, page 26, line 37, at end insert—

“(h) whether the incident involves failure modes not previously observed in the relevant sector materially involving autonomous or adaptive systems based on machine learning, including where the potential impact of such failure modes was mitigated or prevented.”

Amendment 6, in clause 15, page 27, line 7, at end insert—

“(ea) where the incident was associated with one or more autonomous or adaptive systems based on machine learning, details of those systems and their involvement in the incident;”

Amendment 7, in clause 15, page 30, line 8, at end insert—

“(fa) whether the incident involves failure modes not previously observed in the relevant sector materially involving autonomous or adaptive systems based on machine learning, including where the potential impact of such failure modes was mitigated or prevented;”

Amendment 8, in clause 15, page 30, line 21, at end insert—

“(ea) where the incident was associated with one or more autonomous or adaptive systems based on machine learning, details of those systems and their involvement in the incident;”

Amendment 9, in clause 18, page 40, line 10, at end insert—

“(8A) Where the CSIRT receives notification of an incident under regulation 11, 11A, 12A, or 14E that materially involves autonomous or adaptive systems based on machine learning, the CSIRT must share relevant technical information with the relevant body within 72 hours.

(8B) For the purposes of this regulation, a “relevant body” means the AI Security Institute or any successor or replacement body designated by the Secretary of State.”

I will speak to the amendments tabled by Iqbal Mohamed, but wait for the next group to speak to clauses 15 and 16 and the amendments to them in the name of the official Opposition.

From the outset, it is important for me to say that while I have spoken to the hon. Member more generally and responded to a debate he secured on AI, I have not spoken to him specifically regarding these amendments and their precise purpose. However, given his concerns about the AI sector and his background, we can see where he is going with them. Broadly speaking, the amendments would ensure that as part of the reporting requirements under these clauses, there is an ability to measure whether adaptive AI or large language models have been responsible for a cyber-security breach or an incident within the systems themselves.

That derives from what we see happening more generally in the cyber-security sector. We heard evidence that, online, people can essentially purchase a cyber-security hack suite of software. It is possible to pay for people to do hacking and one can get training in it. A lot of hacking and cyber-security breaches are now expanding because of large language models and the use of AI to probe systems. I do not know if we have a sense of scope regarding how much this is a problem specifically in the UK, whether for the individual businesses or organisations that will be regulated under the Bill. I understand, as I interpret them, that the point of the amendments is to get a dataset on where AI or automated decision making has been used to pose a particular cyber-security risk.

The amendments also speak to a more general point. There has been a lot of debate in this place over the years about what we as a country, and equivalent democracies, are doing on the regulation of AI and large language models, building on the Bletchley conferences, innovative work and what guardrails we need to think about in terms of imposing LLMs and AI in the UK, and how we approach AI being used by hostile state actors, such as through bot accounts. I understand that the use of deepfakes, bots and so on is an emerging risk as a method of cyber-attack. There are broader issues with regard to transparency when bots on the internet and social media networks can get into various IT systems and accounts, and effectively pretend to be somebody else to get around the cyber-security system. As with all things, we do not know what we do not know. I understand that the amendments were tabled to increase reporting requirements and give us more evidence of the scope of the problem and the threat posed.

I will be grateful if the Minister gives his sense of how much of a problem this is, particularly with regard to whether reporting requirements are necessary. I believe that the Government’s original plan was to introduce an AI Bill. That would have pros and cons, and I remain agnostic on that, but, speaking for His Majesty’s Opposition, I would like to know the Minister’s plans for the AI landscape and whether, in the upcoming King’s Speech, there is an idea of revisiting an AI Bill, which might make such amendments obsolete.

Order. That is not relevant now.

I appreciate the intent behind the amendments and the Shadow Minister’s position of understanding but not supporting them, which I share. I share his concerns about the potential for emerging risks posed by AI systems, not least in the realm of cyber-security. At the same time, I am conscious that we have not specified any risk factors in the Bill from a reporting point of view for the National Cyber Security Centre or the regulators. To do so in this context would place an undue priority on one category or source of risk.

For those reasons, although I understand the motivation behind the amendments and I am conscious of the risks posed by AI systems, I urge the hon. Member not to press them. The Bill is technology-agnostic rather than focused on particular areas of risk. The Government continue to work on mitigating AI risks, primarily at the point of use, but also through extensive Government capability, not least in the AI Security Institute.

I beg to ask leave to withdraw the Amendment.

Amendment, by leave, withdrawn.

Ordered, That further consideration be now adjourned. —(Taiwo Owatemi.)

Adjourned till Tuesday 10 February at twenty-five minutes past Nine o’clock.

CSRB21 BCS Chartered Institute for IT

CSRB22 Internet Services Providers’ Association (ISPA)

CSRB23 The ABI

CSRB24 Dr Áine MacDermott, Liverpool John Moores University

CSRB25 Rob Wright, Chief Commercial Officer, Hexiosec, Ambassador for Software Security for DSIT

CSRB26 Online Safety Act Network

CSRB27 Shoosmiths LLP

CSRB28 British Insurance Brokers’ Association (BIBA)