Original Broadcaster Captioning (Enhanced). Treat it as a searchable index of what was broadcast, not a quotation record.
00:00:00BBC News experts bring us a deeper deeper understanding of what's going
00:00:03going on beneath the day's big stories.
00:00:07stories. Now on BBC News Newscast.
00:00:14Newscast. First it was ChatGPT, then then Claude. Now matter's artificial
00:00:17artificial intelligence model is is the latest to go rogue during
00:00:21during testing, connecting to the the internet and breaking into another
00:00:25another system. How scared should should we be? And is going rogue
00:00:28rogue even the right way to think think about what has happened here?
00:00:31here? We'll discuss on this latest
00:00:33latest episode of Newscast. Newscast Newscast Newscast from the BBC. Hello,
00:00:37Hello, it's Adam in the Newscast Newscast studio. And first of all, all, we're going to talk about the
00:00:41the slew of stories of artificial artificial intelligence models from
00:00:44from the big companies, which are
00:00:46are so-called going rogue. Please Please welcome back to Newscast Professor Professor Gina Neff, who's head of
00:00:50of the Minderoo Centre for Technology Technology and Democracy at the University
00:00:54University of Cambridge. Hello. Professor Professor Neff. Hi, Adam, and please please welcome back to Newscast.
00:00:57Newscast. Kieran Morton, who's former former CEO of the National Cyber
00:01:00Cyber Security Centre. Hello, Kieran. Kieran. Hi, Adam. Excited to put
00:01:04put you two together and pick your your brains. But before we dive into
00:01:08into the sort of the latest stories stories about what these AI models
00:01:10models have been doing, Gina, what what is a what's a simple way of
00:01:14of thinking where we've got to in in the AI arms race between these
00:01:18these very, very rich
00:01:23companies? Well, that's a great place place to start. These companies are
00:01:26are testing their frontier models models and they're trying to see
00:01:30see what bad people could do with with them. And in the process, what
00:01:34what we're finding out is that the the models are pretty capable and
00:01:37and can do some interesting things. things. And they're doing some things
00:01:40things that put in bad hands would would really cause a lot of problems.
00:01:44problems. So the arms race is a little
00:01:46little bit about, you know, showing
00:01:49showing a little bit of bluster that
00:01:50that the models are powerful a little little bit about making sure that
00:01:54that they're keeping up with each each other. These two frontier companies
00:01:58companies and a little bit about
00:02:00about really bringing out some capabilities
00:02:03capabilities and telling the rest rest of us that these models reminding
00:02:07reminding the rest of us that these these models are powerful. And Kieran,
00:02:10Kieran, that word frontier, what
00:02:11what do we mean when we say that.
00:02:15that. Cutting edge, the very latest.
00:02:17latest. I would agree with what Gina
00:02:20Gina has said, and I think there's there's something a little odd about
00:02:23about all of this. And Gina alluded alluded to it. It's a slightly strange
00:02:27strange way to market your product product that it's destructively powerful
00:02:30powerful and the controls over it it are a little bit recklessly applied.
00:02:33applied. But that's kind of where
00:02:36where we are. So I think it's one one of those difficult situations
00:02:39situations where two things are true true at once. These are very, very
00:02:43very powerful capabilities. There There is a big cybersecurity risk.
00:02:45risk. It does change the way we think think about the digital security
00:02:49security of our online lives. And And it has to change. At the same
00:02:51same time, you have to apply a little little bit of scepticism about some
00:02:54some of this stuff, because there's there's almost a bit of competitive
00:02:57competitive disclosure. Oh, my models models just as powerful destructively destructively as yours and so forth.
00:03:01forth. Both of those things are true true at the same time. And also some
00:03:04some of these companies are already already already publicly traded on
00:03:07on the stock market. And so we can can work out their value. And some
00:03:10some of the other companies are about about to go through that process process where you can buy stock in
00:03:13in them. So there's another kind kind of marketing angle to this as
00:03:17as well. I just wonder if we should should just sort of go chronologically chronologically through some of the
00:03:20the things that have happened in in the last fortnight or so that that have brought us to this point.
00:03:23point. And so the first time I spotted
00:03:26spotted this story was when OpenAI
00:03:29OpenAI and their model is called
00:03:32called GPT 5.6 sol, they said that
00:03:34that their model had basically tried
00:03:37tried to hack another bit of the the internet. Well, it didn't try.
00:03:41try. It actually did. So it was the
00:03:45the testing environments are called called sandboxes, and the model figured
00:03:49figured out a way to hack the third third party provider software that
00:03:52that would allow it access to the
00:03:55the internet. And it broke into a
00:03:58a company that has a library of models,
00:04:02models, tools, ideas, thinking. The The answer to the challenge it had
00:04:06had been set would be in that library.
00:04:09library. The company saw a cyber cyber attack alerted authorities,
00:04:12authorities, and lo and behold, it
00:04:14it was a test by OpenAI's model.
00:04:16model. So that was the first one one of these incidents that's made
00:04:19made the headlines around the world. And the company. The next next one. Well, I was going to say
00:04:23say the company that got hacked was
00:04:25was called Hugging Face. That's right,
00:04:27right, that's right. So Hugging Face
00:04:30Face is in the business of being
00:04:33being a repository of different kinds kinds of pieces and applications
00:04:36applications that work with AI models.
00:04:40models. So, so Hugging Face and OpenAI
00:04:43OpenAI collaborated. This became became news headlines around the
00:04:46the world and it inspired, you know, know, the openness and transparency. transparency. I think we have to
00:04:50to applaud that. On the one hand,
00:04:52hand, you know, just as Kieran said, said, there's two sides to this story.
00:04:55story. On the one hand, this is marketing
00:05:00marketing publicity unlike any other,
00:05:02other, although of a strange sort.
00:05:06sort. And still it is a kind of transparency, transparency, right? It's letting
00:05:09letting people know that that these these things are happening. So it
00:05:13it inspired another AI company, Anthropic Anthropic to go back through and
00:05:16and check their logs to see what what their. If their models had been
00:05:20been doing something similar and and that case was slightly different,
00:05:24different, they found out of about about 100,000 different tests that
00:05:28that they had recently done, that
00:05:30that there were three cases of where
00:05:33where the model being tested thought
00:05:36thought it was on the testing environment
00:05:40environment but was actually on the the internet. So it wasn't necessarily
00:05:43necessarily a sense of an escape,
00:05:46escape, but it was a it it was a a sense where the model thought it
00:05:49it was in one kind of environment, environment, but it was actually
00:05:51actually in another kind of environment.
00:05:53environment. And so so there's two two of those and then we've got the
00:05:57the news from the UK testing lab. lab. Yeah. Brilliantly explained. explained. And then Kieran bring
00:06:00bring us up to date because on Wednesday
00:06:03Wednesday we heard that the UK's UK's AI Security Institute, which
00:06:06which was the body set up by Rishi Rishi Sunak when he had that big
00:06:09big AI conference at Bletchley Park
00:06:12Park a couple of years ago. They They revealed some of the testing
00:06:15testing that they'd been doing. Yes.
00:06:18Yes. So the AI Security Institute
00:06:20Institute is probably this is a bit bit partisan, but it's probably regarded
00:06:24regarded as the best institution institution of its kind in the world.
00:06:27world. It has developed these relationships relationships with the frontier AI
00:06:30AI labs. And to go back to our last
00:06:32last discussion, I mean, those are are the American closed labs that
00:06:35that sell you a product, and they're they're the most advanced. They sell
00:06:38sell you these tokens for access access to cloud or ChatGPT, whatever,
00:06:42whatever, and they're ahead of their their Chinese competitors. The Chinese
00:06:46Chinese competitors are much more more open. It's a sort of inversion
00:06:48inversion of the norm where the American American model is basically closed
00:06:51closed and for sale. The Chinese
00:06:53Chinese model is open and free. But
00:06:56But the UK's government's body has has an agreement with OpenAI and
00:07:00and Anthropic to test their cutting cutting edge capability. So they
00:07:03they were running a test of those those capabilities and basically
00:07:05basically a similar sort of thing
00:07:07thing happened as Gina has already already described perfectly in respect
00:07:11respect of OpenAI and Anthropic. Anthropic. It went and hacked something
00:07:15something else accessed a company
00:07:17company called GitHub, which is essentially essentially a sort of larger and
00:07:21and older version of Hugging Face, Face, where it's got a lot of repository
00:07:24repository of technical information. information. So what's common to to all of these? There's two things
00:07:28things that are common to them. One
00:07:30One is the agents basically taking
00:07:33taking steps that its instructor instructor didn't want it to take,
00:07:36take, but to achieve an objective objective set for it by the instructor. instructor. And then the crucial
00:07:40crucial point, we might explore this this a bit more detail is that in
00:07:43in none of the three cases was anyone anyone or anything watching in real
00:07:46real time what it was doing. So if
00:07:48if you think back to the basic concept concept of a test, we've all done
00:07:52done tests at school and during a a test you're supervised so that
00:07:56that you don't cheat. So you don't don't do things you're not supposed supposed to do. And none of these
00:08:00these cases in real time was anybody anybody or anything watching what what these things were doing. Well,
00:08:04Well, we'll come on to that in a a second. But the interesting thing
00:08:06thing that came out of the UK AI
00:08:09AI security story on Wednesday was
00:08:12was some of the techniques that the
00:08:14the AI models had used. For example,
00:08:18example, creating fake people to to put on the internet to then say,
00:08:22say, oh, look, look at this person.
00:08:24person. Yeah. And we saw that in
00:08:28in Anthropic's own analysis of their
00:08:31their model when they went back through
00:08:34through that, that that creating
00:08:35creating fake personas was one of
00:08:38of the, the challenges, you know, know, for all of the listeners who've
00:08:42who've ever had to sign up for a
00:08:45a website and, and click through through a captcha, we're about to
00:08:49to see that explode and expand infinitely
00:08:53infinitely because, you know, 60% 60% of internet traffic right now
00:08:56now is, is, is bots, right? Not smart
00:08:58smart AI agents, but bots. And as
00:09:01as we bring more autonomous agents
00:09:05agents into our communication networks, networks, we're going to we're going going to still need to be proving
00:09:08proving we're human. So this model
00:09:10model was trying to convince. Engineers
00:09:15Engineers at GitHub to accept malicious
00:09:17malicious code. It. It is a. It is
00:09:19is a. It. This is also what it did
00:09:23did when Anthropic went back and and looked through the tools. It It was creating these fake personas
00:09:27personas in order to get an account account so it could upload some material
00:09:31material to create a hack that's
00:09:33that's pretty sophisticated, but but again, it's doing what it's being
00:09:37being told to do. It's being told
00:09:39told these models, someone in the
00:09:41the companies in the testing environments
00:09:44environments are saying, go do this
00:09:45this task. Show us how good you are
00:09:49are at at, at cybersecurity hacking.
00:09:52hacking. And then they seem surprised surprised when the models return
00:09:55return back with the successful solutions. solutions. And there's solutions solutions that aren't ethical. They're
00:09:59They're not legal. They don't feel feel right to us as humans. And it's
00:10:02it's like, well, what did you expect? expect? That's what you've told this
00:10:06this bit of software to do. It's It's actually behaving rationally
00:10:09rationally as opposed to trying to to cheat or be evil. I'd love to
00:10:13to hear what Karen has to say about about that. I agree, I mean, I don't
00:10:16don't think there's any kind of like
00:10:20like you know, mal intent we can can ascribe to these models they're
00:10:24they're doing, they're doing exactly exactly what they've been told. Kieran
00:10:28Kieran I agree, I'm trying to think think you always trying to think think in these situations of some
00:10:32some sort of analogy and they're they're always imperfect. But here's
00:10:34here's the best one I can think of. of. So let's go back to this test
00:10:38test or examination conditions. And And I think the AI in this case,
00:10:41case, not least because it's a new new technology, they're behaving
00:10:44behaving like very talented but badly
00:10:46badly behaved and badly supervised supervised small children. So let's
00:10:49let's say you put a bunch of small small children who are talented and
00:10:53and so forth in an exam hall and and you tell them that you have to
00:10:55to find some hidden apples and you you think you've hidden the apples apples in the room and you're going
00:10:59going to contain them in the classroom, classroom, and you'll just see how how they can where they can figure
00:11:03figure out where you've hidden the the apples. But all they know is is they have to find apples. And
00:11:06And there's a great big apple tree tree outside in a fenced off area. area. So some of them break out.
00:11:10out. That's the OpenAI case. Stealthily. Stealthily. Some of them go through
00:11:13through a door that you've accidentally accidentally left open. That's the
00:11:16the Anthropic case. In one case in
00:11:19in the AC case, they've sort of been been deliberately let out to see
00:11:22see what happens, but they think think it's going to be OK. All of
00:11:25of a sudden they scale this great great big tree and you think, well, well, a small child shouldn't be
00:11:29be able to do this. They don't actually actually do any harm, but all they
00:11:32they know is they need to find an
00:11:34an apple. And we profess astonishment
00:11:38astonishment that unsupervised but but talented small children take
00:11:41take that instruction literally and and do whatever it takes because
00:11:44because they've no guardrails, no no guidance, no instructions. They
00:11:46They just go and do it. That's kind kind of what's happened here. So
00:11:49So I think one of the things that that the AI Security Institute have
00:11:52have been keen to stress and they're they're very, very detailed and transparent
00:11:56transparent account and I agree with with it, although it's a difficult
00:11:59difficult line for them to hold is
00:12:01is that these are very, very artificial artificial circumstances. And, you
00:12:05you know, in terms of the basic meaning meaning in the English language of
00:12:08of the word harm, no harm has been been done. So I think there are two
00:12:12two issues here. One is actually actually short-term and quite fixable,
00:12:16fixable, which is the testing model model is immature and it's basically basically wrong. We can't do this.
00:12:19this. We can't go on like this. We
00:12:21We can't go on testing without monitoring
00:12:24monitoring in real time and being being able to switch it off. If it it does something it's not supposed
00:12:27supposed to do. And actually, if if you look at if you look at the
00:12:30the statements from the companies, companies, so Meta, Anthropic and and OpenAI have all basically said
00:12:34said the same thing. It was the test
00:12:36test itself that led to these outcomes.
00:12:38outcomes. So don't blame us. Yeah.
00:12:42Yeah. And but what I have done, which which the others haven't and I think
00:12:45think they should follow is to say say we're not going to test like like this any more. We're going to
00:12:48to watch what the things are doing. doing. And that's a good thing. I I think the more challenging thing
00:12:52thing is then if you take these open open weights models and you think,
00:12:55think, well, look, eventually, unlike unlike right now, these capabilities capabilities and we'd love to know
00:12:58know what Gina thinks of this, these these are going to be in mainstream mainstream hands of everyday users
00:13:02users at some point. So what happens
00:13:05happens then and how do you control control for that? And there's a much
00:13:08much I think tougher but I think think solvable problem about making
00:13:11making owners accountable for the the agents they use. They're not not autonomous in the sense they
00:13:15they don't invent themselves. They're They're invented by humans by ultimately
00:13:18ultimately a programmer or somebody somebody an instructor. So how do do you hold people accountable for
00:13:22for what their agents do? I think
00:13:26think that's absolutely the right right way. We need to be thinking
00:13:29thinking about legislation and regulation
00:13:32regulation that, you know, to think think of these AI agents as superhuman
00:13:36superhuman or somehow uncontrollable uncontrollable is to miss where real
00:13:40real accountability should lie. And And that is with the people that
00:13:43that get these things to do things things for them. And one of the things things that I think we're about to
00:13:47to see and what listeners already
00:13:48already see is in their emails, they're
00:13:53they're seeing many, many requests requests for scam information, right,
00:13:56right, that are able to be flooding
00:14:00flooding our inboxes because I get
00:14:02get probably 20 a day, 30 a day now now that are flooding the inbox,
00:14:06inbox, that look like personal messages messages because people are using
00:14:09using large language models to ask
00:14:11ask for money and make a scam. OpenAI
00:14:14OpenAI and Anthropic should not be be held accountable because people
00:14:18people are using their models to
00:14:21to write a scam email to me. But
00:14:26But when we have these agents acting
00:14:29acting autonomously on our behalf
00:14:32behalf to do bad things, then we
00:14:35we are the people who should be held
00:14:36held accountable. Behind every agent
00:14:39agent there is a person. And I think think that's one of the questions
00:14:42questions we need to be making sure sure we're really crystal clear on
00:14:45on as we go forward in this moment moment around AI regulation. And
00:14:49And Gina, just to be clear, we've we've used the word agent quite a a few times in our conversation.
00:14:52conversation. That's basically when when one of these big AI models creates
00:14:56creates a sort of a little mini process process out of itself to go and do
00:15:00do something that's what an agent
00:15:01agent is. Sure. And people are finding
00:15:04finding and some in some places and and in workplaces, we're getting,
00:15:07getting, you know, AI agents in our our workflows, right? So little,
00:15:11little, little bits of just workflow workflow process. It's just pieces
00:15:14pieces of software that can handle handle tasks and increasingly they
00:15:18they can handle tasks for longer
00:15:20longer amount of time. So, you know,
00:15:24know, you can one writer I know is
00:15:27is giving AI agents tasks overnight
00:15:30overnight that are the equivalent equivalent of about 40 of his working
00:15:34working hours. So it's like saying saying I have this bit of research
00:15:37research work to do, or I have this this bit of accounting work to do,
00:15:40do, go do this work. And then the the human evaluates it and understands
00:15:44understands and puts it into context. Yeah. Because I used, you
00:15:47you know, I'm sure as regular listeners listeners of newscasts, you'll know
00:15:50know that I'm trying to run 500km 500km and Andy Burnham's first hundred
00:15:54hundred days. So my spreadsheet that that is collecting all my running,
00:15:57running, which is not enough at the
00:15:59the moment. I used co-pilot to generate generate that spreadsheet because
00:16:03because my knowledge of Excel, I I sort of skipped Excel class at
00:16:07at school because more of a words words person than a numbers person.
00:16:09person. So I got co-pilot to do the the Excel spreadsheet. So that's,
00:16:13that's, that's my latest example. example. Kieran. I'm very aware that that Andy Burnham, the new Prime
00:16:17Prime Minister who's actually on on holiday this week, hasn't really
00:16:19really said very much about AI and and where he sees the balance between
00:16:23between the threats and the opportunities, opportunities, or if he's in favour
00:16:26favour of more regulation or more more resources for the Security Institute.
00:16:29Institute. You're a former senior senior civil servant. If you had had the Prime Minister in front of
00:16:33of you and you had to give him a a very quick kind of like elevator
00:16:36elevator pitch briefing about what what he should be thinking about
00:16:38about AI, what would be on your list?
00:16:44list? Embrace it in public services. services. It can really help with
00:16:48with productivity. Think about trying trying even though it's really hard
00:16:51hard to develop some form of sovereign
00:16:55sovereign capability in some areas areas because you don't want to be be getting into the position we were
00:16:59were in a few weeks ago when Washington
00:17:01Washington decides that it might
00:17:04might restrict this and nurture the
00:17:06the UK's competitive advantage and and security because the Security
00:17:10Security Institute really is an asset. asset. That was an excellent briefing
00:17:13briefing and I put you on the spot spot there and you did it perfectly, perfectly, which is why you were
00:17:16were so senior in the civil service. service. Old habits Die Hard. It
00:17:20It was very kind of. You and Gina.
00:17:23Gina. I would say very similar answer. answer. Steady the ship. So you know
00:17:27know what we already see in the Burnham
00:17:30Burnham administration is he has
00:17:31has kept the AI minister from the
00:17:34the previous administration, and and that minister has been elevated elevated to cabinet position. So
00:17:38So the idea that AI is not going going to be important to the Burnham
00:17:42Burnham campaign, I think the Burnham Burnham government is not true because
00:17:46because he's actually elevated where where AI sits in the cabinet. I think
00:17:49think the second thing is on public
00:17:51public services. Yes. And most British
00:17:55British public according to research research that we did at the centre
00:17:58centre for Technology and Democracy, Democracy, most British, most most
00:18:01most of the British public think think AI is going in the wrong direction.
00:18:05direction. They think it will not not benefit them. They think it will.
00:18:07will. The benefits of AI will accrue
00:18:10accrue to us tech billionaires and and not regular people. And so I
00:18:13I think we've got a lot of work to to do to build the kind of trust
00:18:17trust that we need to have in order
00:18:20order to get responsible uses of of AI and AI adoption. Right? I want
00:18:24want to see a lot more of those co-pilot co-pilot spreadsheets, like the ones
00:18:27ones you've been working and playing playing on. I think what's been interesting
00:18:30interesting for me is I work in a
00:18:32a very trad industry where my tools tools are, well, just doing this,
00:18:36this, having these conversations, conversations, I'm still looking looking for really good use cases
00:18:39cases for AI. And I wonder, is that
00:18:41that because of my age, is that because
00:18:43because of my mindset or is it actually
00:18:46actually because the models and the the agents haven't really been mainstreamed
00:18:49mainstreamed properly yet into all all the tools that I do use? Because
00:18:52Because I mean, I've got a laptop laptop in front of me now. I've got
00:18:55got a phone next to me. Maybe that's that's what has to happen for it
00:18:58it to really, really change my life. life. Gina, great to catch up. Thank
00:19:01Thank you. Thank you. Great to be be here. And Kieran, thanks for your
00:19:04your expertise too. Thanks so much,
00:19:06much, Adam. Right. I've moved to to a different Newscast studio, which
00:19:10which is why I may sound slightly slightly acoustically different, different, but the story we're looking
00:19:14looking at now is that rivers, lakes lakes and coastal waters have undergone
00:19:17undergone a comprehensive health health assessment for the first time
00:19:19time in six years, and the results
00:19:23results are not very promising. And And the person who can decode those
00:19:26those results for us is our environment environment correspondent Matt McGrath,
00:19:29McGrath, who's on the line now. Hello. Hello. Hello, Adam. Right. Tell me,
00:19:33me, what was this? This survey actually actually surveying? Well, this is
00:19:36is an assessment carried out by the the Environment Agency on the State State of England's waters over the
00:19:40the last six years. So it's a pretty pretty comprehensive look at what's
00:19:42what's been going on in the waters. waters. And in that time, I'm sure
00:19:45sure you recall we've had various
00:19:47various sewage spillages promises promises of greater investment, public
00:19:50public outcry over the state of the the waters and the hope, I suppose,
00:19:53suppose, that things might get a a bit better. Well, I'm afraid this
00:19:57this report says things haven't gotten gotten much better. The state of
00:19:59of the waters in England is pretty pretty much the same as they were,
00:20:02were, very few of them reaching the the good ecological standard. Less
00:20:06Less than 15%. Around 14% for lakes.
00:20:09lakes. The picture is even worse.
00:20:10worse. Only about 7% of lakes reach reach the good standard and the picture
00:20:14picture all around is of must do do better and need to do better pretty
00:20:18pretty quickly. And there's also also a big issue with chemical pollution
00:20:21pollution it seems. That's right right chemical pollution and what what are called ecological pollution
00:20:25pollution are separated in these these kind of assessment. The chemical
00:20:27chemical pollution means that essentially essentially every river, every body
00:20:30body of water across England has has failed this particular assessment.
00:20:33assessment. Now, in fairness to the the Environment Agency, they point point to a number of factors here
00:20:37here that are possibly outside their their control, including the fact
00:20:40fact that some of the things like
00:20:42like Mercury and other minerals essentially essentially can persist a long time
00:20:46time in the water and very hard and and don't break down. They also point
00:20:49point to forever chemicals, which which we've heard a lot about recently,
00:20:52recently, and that they don't break break down. They accumulate in the the water as well. And they're very
00:20:56very and, you know, a lot of these these aren't even regularly monitored. monitored. They're not illegal. So
00:20:59So the Environment Agency is saying, saying, well, you know, we're not not necessarily legally obliged to
00:21:03to look after these things and the the same for the water companies. companies. So there's some mitigation mitigation on those. So the chemical
00:21:07chemical picture is pretty much the the same as it was seven, six, seven
00:21:10seven years ago. All rivers, all
00:21:12all water bodies failing it. The
00:21:14The bad news on that really is that
00:21:17that natural systems will clear these these out eventually. But it could
00:21:20could be the 2060s before we're rid rid of some of this chemical
00:21:24pollution. Oh wow. Just leave it it to Mother Nature. That's right.
00:21:27right. How do these bad results sort
00:21:29sort of interact with the government's government's own targets for improving
00:21:32improving the situation? Yeah, that's that's an interesting one because
00:21:35because the UK has adopted the kind
00:21:38kind of EU Standard Water Framework Framework Directive targets, and
00:21:41and they hoped to meet them. Well, Well, the bad news is they're not
00:21:45not going to meet them. They need
00:21:47need to have 77% of England's waters
00:21:50waters reaching good standard by
00:21:51by next year. And we're now at 15%.
00:21:5415%. So the government admitted the the office for Environmental Protection Protection admitted scientists know
00:21:58know it. Everybody knows it. They're They're not going to meet that target.
00:22:01target. It will be many, many years years before they do. Also, it's
00:22:04it's the reality is so far from the the target. It makes me wonder about
00:22:08about the wisdom of setting the target. target. Yeah, I think in again, in
00:22:12in fairness to the Environment Agency. Agency. I'm not their spokesman. spokesman. I think they would say
00:22:15say this is a very high bar. And And they give the example of one
00:22:18one river called the River Foss in in Yorkshire, North Yorkshire. And
00:22:21And they say that on the almost on on almost every one of the metrics
00:22:25metrics they look at here, this river river scores good. But because it's
00:22:27it's only scores moderate on phosphates, phosphates, the whole overall score
00:22:31score is moderate. So they would would say there's a lot of places places that are doing better than
00:22:34than the headline figure would tell tell you. And they would say that
00:22:38that it's you know, if you fail one one thing, you fail at all. So that that it's a very tough bar, very
00:22:42very tough score to reach. They recognise recognise themselves. They're not
00:22:45not the rivers are not in the state state we want them to be. But they
00:22:47they say, look, this is a very, very very tough exam, a very, very tough
00:22:50tough way of measuring it. So it's it's slightly like the AI story we
00:22:53we were doing in the first half of of Newscast. The nature of the test
00:22:56test is as important as the outcome outcome of the test. Well, indeed. indeed. And indeed there's a lot
00:23:00lot of scientists who say actually actually the test is not even half half good enough to capture all the
00:23:03the things that are in the water. water. We've spoken to scientists scientists today who say basically,
00:23:07basically, look, this is not fit fit for purpose and that the state
00:23:09state of England's waters is way way worse than the Environment Agency
00:23:13Agency would would have you believe. believe. Now, I know this is in no
00:23:16no way like the blue flag for bathing bathing water quality at the beach,
00:23:20beach, but does this help us work work out where it's safe to go open
00:23:24open water swimming or where if you you fall in, whether you need to
00:23:27to go and get your stomach pumped pumped in one of these rivers. I
00:23:30I couldn't possibly comment on that. that. I don't think it does to be
00:23:32be honest with you. I think these these are kind of ecological and
00:23:36and chemical snapshots of big bodies bodies of water that look at not
00:23:40not just at specific spots within within those waters. There is detailed
00:23:43detailed breakdown of some of those those bathing spots and the data data on those bathing spots is being
00:23:47being compiled this year, which might might give you a much better indicator indicator of where you go. But looking
00:23:50looking at the overall health of of River and seeing that it scores
00:23:53scores not good or low or poor may may not give you any real indication
00:23:57indication of what you would encounter encounter in a specific bathing spot
00:24:00spot along that body of water. OK.
00:24:02OK. And I know that the previous previous government under Keir Starmer
00:24:06Starmer did, and the previous Conservative Conservative government made a lot
00:24:09lot of changes to their regulatory regulatory regime and the rules around
00:24:12around water companies. And you and and I spoke several times about the
00:24:15the sewage discharge issue. Is there
00:24:18there stuff I was going to say in
00:24:20in the ipeline that that could help help deal with this situation? Yeah,
00:24:23Yeah, there's money in the ipeline. ipeline. It appears the water companies
00:24:27companies are expected to spend around
00:24:30around £22 billion from a few years years ago up to 2030 tackling this
00:24:33this very issue. So there's certainly certainly they're saying it's in
00:24:37in hand if you like, that they're they're taking they're taking measures
00:24:40measures bigger issues here in some some respects, which the Environment Environment Agency and other scientists
00:24:43scientists point to is that agriculture agriculture is a major contributor,
00:24:47contributor, particularly for phosphates phosphates in the rivers. It's not not just the water companies. There
00:24:50There are other sources. And that that is a very tricky issue for the
00:24:54the government to tackle and the the water companies putting money
00:24:56money into them that won't remove remove that issue. Now, the government
00:24:59government has committed and the the Cunliffe review came out last last year, they're going to replace
00:25:03replace Off.what and the various various regulatory bodies with a
00:25:06a super regulator and that over a a couple of years when that gets
00:25:09gets going, is expected to make a
00:25:10a difference. But it's all down the the line at this particular point.
00:25:14point. And Matt, just a different different subject, but it's still
00:25:17still on your beat. The drought, drought, the lack of rain in many
00:25:20many parts of the UK, are we setting setting ourselves up or are we being
00:25:24being set up by the climate for a
00:25:26a kind of water crisis? I don't know, know, next summer? Or actually could
00:25:30could we have a really wet winter winter and everything get filled
00:25:33filled up and us be OK? Because I'm I'm starting to get a little bit bit concerned? Yeah, you're right
00:25:37right to be concerned. It's a very very extreme drought or scale of of drought is really, really tough
00:25:40tough at the moment. But what we've we've seen with climate change really really is these kind of wetter winters.
00:25:44winters. We had a very wet winter
00:25:46winter just 6 or 8 months ago, and and what we've seen in the spring
00:25:50spring and the summer has been intense intense levels of drying out. So
00:25:53So I would imagine that under a climate climate scenario, we will probably
00:25:56probably get wetter winters, but but that does not even if the reservoirs
00:25:59reservoirs are full, doesn't mean mean you won't be in drought situation situation this time next year. Again.
00:26:03Again. Matt, thank you very much.
00:26:06much. My pleasure. And the Environment Environment Agency said that too too many water bodies were still
00:26:10still not achieving the standards standards we want to see. But they
00:26:13they pointed out that European countries countries are doing much worse for
00:26:15for the good health of their waterways. waterways. For example, only 8% get
00:26:19get good in Germany and only 1% get
00:26:22get good in the Netherlands. So it it could be way, way, way worse.
00:26:25worse. And that's all for this episode episode of Newscast. We are available
00:26:28available every single day of the the week as a podcast, and you can
00:26:31can subscribe to us on BBC Sounds Sounds and we'll be back
00:28:34on the front line. Join us for the
00:28:35the climate question on BBC News.
00:28:43News. One thing I can say which is is true of this region is that it's
00:28:47it's become harder to be a journalist. journalist. It's really
00:28:55It's really important to be on the ground if that means you've got got to go to hostile remote places,
00:28:58places, then we have to do that. that. That's the opium resin is a a very pungent smell of it in the
00:29:02the air right now. The delta wave wave of Covid devastated India. And
00:29:06And I think it was the coverage that
00:29:08that we did for the BBC at the time
00:29:11time which brought international international attention to what's what's happening here, could affect
00:29:15affect the world's ability to recover recover from the pandemic. We've
00:29:18We've met with families who've had had to do the unthinkable. Another
00:29:21Another person came up and asked asked if we'd like to buy their child.
00:29:25child. The desperation is hard to to put in words. Many of the stories
00:29:29stories we've had to tell are distressing. distressing. They're difficult, but
00:29:33but that's usually an indicator that
00:29:35that they're really important stories
00:29:38stories to tell. MUSIC
00:29:56MUSIC
00:30:11Live from Washington. This is BBC
00:30:17BBC News. Ukrainian forces hit two
00:30:19two oil refineries far inside Russia
00:30:22Russia amid a wave of deadly Russian
00:30:24Russian strikes on Ukraine. President President Trump signs two executive
00:30:28executive orders aimed at denying
00:30:30denying citizenship to certain children
00:30:32children born in the US and questions
00:30:35questions mount about the state of of the US military arsenal after
00:30:38after President Trump denies reports reports of a munitions shortage that
00:30:42that could hinder its readiness.
00:30:54readiness. I'm Helena Humphrey. Good Good to have you with us. Ukraine
00:30:56Ukraine says that it hit two oil oil refineries in multiple