This moment, Alex's mile swim, it means everything to him and to all of us.
At Evernorth Specialty Services, we help providers make sure patient care doesn't get interrupted, especially for rare, chronic and complex conditions.
Now Alex and his care team have his hemophilia under control with life-saving medication and care management from Evernorth's specialty pharmacy Acredo.
So Alex can be the last one out of the pool.
Evernorth Specialty Services, because every moment counts.
Visit evernorth.com slash specialty to learn more.
AI had the time of my life.
AI never failed this way before.
From building timelines to assigning the right people and even spotting risks across dozens of projects.
Monday Sidekick knows your business, thinks ahead and takes action.
One click on the star and consider it done.
And I owe it all to you.
Try Monday Sidekick, AI you'll love to use, on Monday.com.
Welcome back to The Nature Podcast.
This week, identifying how a tick-borne virus gets into the brain, and Nature's Bot that's been trained to look at how US research grants are being cancelled.
I'm Nick Perchichow.
And I'm Benjamin Thompson.
A future treatment for a very unpleasant disease may be a step closer, thanks to a new study published in Nature this week.
And it's a disease that can be caught by folk who work outdoors or who, like me, like to go for long walks in the woods, as it's spread by tiny unwelcome visitors that can be picked up along the way.
I'm talking, of course, about ticks, tiny blood-sucking arthropods that are found in wooded areas around the world and live by feeding off the blood of other animals.
But while most tick bites are ultimately harmless, if a little painful, in rare cases these animals can carry diseases that are transmitted by their bites.
One of these is called tick-borne encephalitis, or TBE, caused by the tick-borne encephalitis virus.
It's estimated that there are around 10000 cases per year of TBE across Europe and parts of Asia, and numbers are increasing.
Most people who get an infection have no symptoms, but some experience a short spell of flu-like illness that goes away.
An unlucky few get a few weeks' respite before experiencing a second stage of infection that can cause incredibly serious neurological symptoms.
While an effective vaccine to prevent tick-borne encephalitis is available, its use in endemic countries is low.
To date, there are no treatments available.
But the new paper may be changing that by providing insights into the way the TBE virus interacts with the human body.
One of the team is Sara Gredmark-Roos from the Karolinska Institute in Sweden.
I called her up to find out more about the findings.
Before then, she explained how the second stage of a TBE infection can present itself.
During this phase, you will develop high fevers, usually headache.
You can develop different neurological symptoms such as confusion, dizziness.
You can also develop paralysis.
And this is a time when people usually seek medical care.
And the infection can be of different severities.
And in very rare cases, it's possible to die from the infection.
And after this second phase, up to 40 of the patients will actually develop long-lasting symptoms that will last for years.
It could be concentration problems, fatigue problems, or if you had a paralysis that can stay.
And what was not known about this virus then?
And what questions were you looking to answer with this work?
So it wasn't known how the virus gets into the cell.
It's also not known how the virus gets into the central nervous system.
So, as a start, we wanted to know how does the virus get into the cell to be able to further understand the disease mechanisms and then potentially also be able to find a treatment for the disease.
Since TBE has no known treatments, this was a question that we wanted to find an answer to.
This is what you and your colleagues have tried to figure out then.
And to go about it, you've used human cells in a dish.
And each of these has a different gene inactivated.
Knocked out is the technical term.
And you've done that using CRISPR-Cas9, to see which of these genes encode proteins that the virus needs for its replication process.
It seems that you found a fair few, but there was one in particular that I think is the main focus of your paper, and this is a gene called LRP8, which encodes a protein called LRP8.
That's correct.
So we found LRP8 that stood out.
And the reason why we wanted to study it further is because it is a cell surface protein.
LRP8 is then a lipoprotein receptor.
And lipoprotein receptors are a big family.
And there is also some other lipoprotein receptors that's been shown to be important for other virus infections.
And am I right to say that this one is found a lot on the surface of brain cells?
Yes, so that was also one thing that was very exciting when we first saw our data from this first experiment is that allopate is highly expressed in the central nervous system, so in the brain cells, in neurons and also in astrocytes.
So in your paper then, you've put forward evidence that the TBE virus attaches to this receptor, this surface protein, which has a normal job, but in this case the virus is hijacking it to in some way get inside cells.
Yeah, so we could show that the virus attach to the protein and then it gets internalized together with the protein.
So once we had this potential candidate that is important for TBE infection, we knocked out the LRP8 gene and we could then see that we greatly reduced virus infection in those cells.
And instead, if it overexpressed the protein, we would instead see a much higher infection rate.
You also show that it's possible to stop this happening then, by essentially bunging up the virus's ability to bind to the cell by using synthetic parts of the receptor to get in the way.
Yes, that's correct.
And in order to do that, we had first narrowed down what part of the receptor was important for the TBE binding.
And then instead, we expressed that part and incubated the virus together with this binding part, and then we were able to show that it binds and thereby reduce the infection in vitro and in vivo.
Speaking of that, then a lot of the experiments you did involved non-brain cells, but you also tested this out using neuronal cells derived from stem cells and in a mouse model as well.
What did you see there?
Yeah, as for the mouse experiments, we pre-treated the virus with the necessary part of the receptor, or pre-treated mice with the necessary part of the receptor.
And we could then see that these mice were protected from clinical disease.
You've shown evidence then that this LRP8 receptor plays a key role in infection by the TBE virus.
Obviously, this is one member of a group of viruses that include things like Zika and West Nile, viruses carried by mosquitoes, and there are other tick-borne ones too.
Does this work show that this particular protein could be important there as well?
It seems like LRP8 is used by tick-borne encephalitis but not by other flaviviruses.
We performed experiments using a number of different flaviviruses and could not see the same effect, and we of course don't know what this means.
But As for TBE, it is a virus that do infect the brain and that gives encephalitic symptoms.
So it might be that LRP is important in the particular pathogenesis of TBE and encephalitis.
And I wonder what this means, because there are vaccines that can prevent people getting TBE.
And you yourself are an infectious disease clinician.
What do you think this result might mean more broadly?
So vaccine coverage is not very good in big parts of Europe and other parts of the world.
There is no treatment to tick-borne encephalitis virus infection.
So we are hoping, by the finding of this receptor, that this can help us in identifying drugs that could either work on inhibiting the virus binding to the receptor or maybe inhibiting downstream effects from the activation of the receptor.
Coming up, nature has built a bot to figure out how grants are getting cancelled in the US.
First though, Dan Fox is here with this week's research highlights.
A robot that squirms its way down your windpipe may sound like something from a horror film, but this new device could save lives.
That's because even extensively trained emergency responders sometimes struggle with quickly inserting air-supplying tubes into their patients' throats.
So to solve this problem, researchers have designed a robot that autonomously guides a soft plastic tube down the throat and into the windpipe.
After only 5 minutes of training.
Emergency providers using this robot method on cadavers were, on average, 24 seconds faster than if they used a video-guided method.
They also succeeded on their first try 9 out of 10 times, compared with less than two thirds of the time using the video technique.
These results suggest the robot could help to make the life-saving procedure more accurate, reliable and accessible to a wider range of medical personnel.
Read that research in full in Science Translational Medicine.
10000-year-old bodies found buried in a crouching position might be the world's oldest intentionally preserved mummies.
Crouched burials occurred across Southeast Asia between 14,000 and 4,000 years ago.
These remains resemble bodies that are preserved in a seated position by some modern peoples in the New Guinea highlands, who use smoke to dry the remains of important community figures.
But it's possible that some ancient remains could bear this resemblance through processes like cremation.
To learn whether these ancient remains were mummified.
Researchers used techniques to detect signs of heating in 54 prehistoric skeletons unearthed in China, Vietnam and Indonesia.
The researchers found that many of the skeletons had been slowly heated over a long period, suggesting they had been deliberately smoked.
Given the resemblance to modern practices, it's likely that these bodies were mummified by tying them up and hanging them over a smouldering fire for as long as three months.
The oldest of these crouched burials showing evidence of smoke drying predate the earliest known mummies from ancient Egypt and other cultures by several millennia.
Unearth that paper in the Proceedings of the National Academy of Sciences of the United States of America.
This week the Nature's Career section have been doing some experimenting of their own and using machine learning to explore an alternate timeline.
As you've probably heard on the podcast before, Nature has been reporting this year on the Trump team's science funding cuts.
Thousands of research grants have been cancelled in the US, although the government haven't revealed exactly how they decide which projects get the chop.
Of course, we can never truly know what discoveries might have been made if this research hadn't been cancelled.
Crystal balls don't exist, after all.
But this week some of our colleagues have been looking to the past to try and get a sense of what the future might hold.
They've trained a machine learning model on the decisions made by the Trump team and then let it loose on grants from 10 years ago to see what might have been cancelled and what the effects on science may have been.
So, to talk me through all this, I'm joined by Jack Leeming, Chief Editor of the Careers section of Nature.
Jack, how's it going?
Very well, thank you.
How are you?
I'm good, thank you.
Thank you so much for joining me to talk about this.
So Jack, you and your team have been building this model to look at grants.
Tell me how it works.
The way it works is.
We've trained a machine learning model to look at the properties of cancelled grants.
And then we've asked it to look at grants in the past to understand whether or not those would have been cancelled.
How was this bot built?
What are the things it's looking for to see what might have caused the grant to be cancelled?
Yeah, so first of all I should credit my colleague, Vera Ninaba, who did all the really clever stuff here.
She's the data scientist.
I'm just a reporter.
But she trained this bot to look at the amount of money left in a grant, to look at individual keywords within a grant, to look at the sort of time period of the grant, how long it might last and other sort of metadata, sort of properties of a grant.
I think the really important thing is the keywords.
So one small piece of evidence we have of what the NIH might be doing is a list of 200 banned words and phrases circulated by the Trump administration to lots of different agencies working within the US government.
And those band word and phrases include things like diversity, transgender.
And the idea is to just remove those words from being mentioned on any US government stuff.
So whether that's kind of press releases, websites, whatever.
And so we use that maybe as inspiration for this bot.
So we looked at those 200 band words and then asked the bot to identify other band words that might trigger cancellation.
And we found a sort of similar list, right?
So among the sort of phrases and words that model found were most likely to lead to grant cancellations were things like gender affirming care assigned male at birth, racial justice, LGBTQ and hate speech, and also word diversity.
And so you've built this model, and then, to see how accurate it was, you compared what it thought might be cancelled to what was actually cancelled.
How did it do?
It was around 70% accurate.
So after we deleted the training data, our model could identify a cancelled grant with 70% accuracy.
Grants that have been cancelled now rather than 10 years ago.
That's right, yeah.
And then the second step to this is... is we then set that bot loose on historic grants.
So something that's really difficult about this piece about understanding the kind of destruction that's happening to US science is you don't really know when you're taking away grants.
You don't really know what the impacts will be.
You will never see what size those grants might have produced.
Yeah, you don't know what the future holds.
Exactly.
So we applied this bot to grants for active 10 years ago because then we could look at the science that came out of those grants, those 10-year-old grants.
And when we did that and looked at the science and sort of ranked it by the citations of a number of papers that might have never existed, what we found is it's so destructive in lots of ways you don't really expect.
So, among the sort of most highly cited pieces of science that we identified, that would have been cancelled had this happened 10 years ago, where things like DropSeek, which is a method for doing lots and lots of RNA processing really really quickly and effectively, that wouldn't have existed.
And you know that's got nothing to do with gender-affirming care, transgender health diversity, and yet it ended up being cancelled anyway.
And what may have caused that to have been cancelled?
What were the things that got the bot to say okay, this one probably wouldn't have made it sort of medical doctor training program?
And lots of those have been cancelled.
So we think that's sort of part of it.
There's also lots of mentions to diversity, to other words with kind of dual meanings.
You know, diversity is a really interesting one because you can have genetic diversity.
And so do we know that.
You know, you said there's some evidence that there are these 200 banned words.
Do we know that this is how the Trump administration is actually cancelling grants?
No, we don't.
We don't know exactly how the administration are identifying grants and how they're actually targeting them.
You know, the 200 banned words, I think, is a really good start.
But we don't really know how they're doing it.
And that's a real problem, right?
I mean, you know, the American taxpayer funds science and has a right to understand what's being cancelled, for what reasons, under what kind of circumstances.
And then some other clues are around from the testimony of someone who spoke to ProPublica.
That person was working for Doge, the Department of Government Efficiency.
He set an AI loose on veterans affairs contracts and just asked the AI to classify it as kind of cancelable or not.
You know, is this an essential thing?
So we know that they're using machine intelligence in this way to cancel some contracts in other departments.
And so I don't think it's too much of a leap of faith to think the NIH might be doing something similar.
And you mentioned drop-seek here.
What were some of the other things that surprised you that would have been cancelled using these criteria?
Yeah, I mean according to our model.
I mean, the most striking one was the human microbiome project.
Oh, wow.
Yeah, we've heard of it.
That's a big old piece of science.
And, according to our model, it identified grants within the human microbiome project that wouldn't have existed.
And therefore cancellation ultimately of the entire project.
And that's an entire start of science.
That's an entire sort of field, a new medical field that yeah, it's like a foundational study exactly that might not have existed, had again an event like this happened 10 years ago.
And another one that surprised me when I was reading through this story was there's a lung cancer treatment that may not have existed had this happened.
Yeah, that was a really big, important medical finding.
So our bot identified the funding for a big hospital and sort of the funding for a massive medical study around lung cancer at that hospital.
And it's a really important sort of medically significant cohort.
They did an awful lot of good work.
And yeah, that wouldn't have existed either.
And you actually reached out to some of the researchers whose grants would have been cancelled with this criteria.
What did they have to say?
Yeah, it's a really funny one.
I mean, I briefly worked at the opinion team at Nature and one of the things I was working on was obituaries.
And it kind of reminded me of that, because reaching out to a researcher asking them to write an obituary for a dear friend, I'm sure you understand, needs to be done with real care.
But the very first line in your email must start with, this person isn't actually dead.
Similarly, when I was emailing all of these researchers and telling them their grant was cancelled, I didn't need to make it really clear like your grant isn't actually cancelled.
This is a theoretical thought experiment.
We trained a robot and it found your grant would have been cancelled.
And people, I think, found that very funny.
I interviewed one person recently We sort of retrained the model with extra data when we got it.
And it turned out, in between sort of booking to interview him and actually talking to him, his grant was sort of reinstated.
It wasn't actually cancelled.
He was really disappointed.
Disappointed?
Yeah, he was like, he really wanted to talk about his cancelled grant.
He was like, oh, no, I've survived.
Which was funny.
And I guess that's another thing to consider as well.
Like the Trump team, seems to be targeting specific kinds of areas, as you say, transgender health and that sort of thing.
But it seems like the criteria they're using aren't really achieving that.
Yeah, what we've done reflects the Trump team's actions.
But really, this isn't a good way to cancel grants.
Politicians and elected representatives of the people have a right to decide where public money goes.
That's ultimately what scientific grants are.
But I think they also need to decide that wisely.
So this is a model looking at past grants, cancelling them, even though that never really happened.
Is there a lot that we can actually say from this?
No.
This model is trained on the actions of the NIH in recent years and the Trump administration.
We don't know had Trump and this shift in culture happened 10 years ago.
We don't know what they would have done.
And it's taking place in a different context.
But all we can say is that this model has been trained on their actions now.
And this is what it's found.
Nature's Jack Leeming there.
For more on that story and the rest of our coverage on the ongoing situation in the US, check out our show notes for some links.
Finally on the show, it's time for the briefing chat, where we discuss some highlights from the Nature Briefing, which is, of course, Nature's daily science newsletter.
Nick, why don't you go first this week?
What have you been reading about?
So I was reading an article in Nature all about AI psychosis, which you know, you may have heard of.
It's been in the news quite a bit recently.
And what I'll say before we get into this is... is that there has been very little research on this.
So this article has been trying to understand what we do know about this and the questions yet to answer.
And when we say AI psychosis, then we're not talking about the AIs themselves.
We're talking about the users.
Yes, that's right.
We're talking about the users.
And so psychosis is when people have trouble distinguishing what is real from what is not real.
They may have hallucinations, they may have delusions and it's often associated with mental health disorders like schizophrenia, bipolar disorder and that sort of thing.
But there have been reports of people getting psychosis from interacting with chatbots.
So You know, they'll be discussing something with ChapGBT or something, and that triggers psychosis.
So are these folk then who have experienced psychosis and the AI is exacerbating it?
Or is this a situation where AI is leading to psychosis?
Well, we don't know is the short answer.
Whether or not AI can trigger psychosis is still a hypothesis.
There have been some reports of people developing psychosis after interacting with these.
There was a preprint that went online last month, so a study that's not been yet peer-reviewed, that said at least 17 people have been reported to have developed psychosis after interacting with chatbots.
But whether one leads to the other is not clear.
And what researchers do know is that it seems that only people that have a predisposition for psychosis if they have schizophrenia or something like that, or they have paranoid tendencies then it's possible that AIs could lead to psychosis.
But it's not quite clear if there's a direct link yet.
But if you don't have a condition such as that, then it seems that your risk of developing psychosis is the same whether or not you interact with a chatbot.
Right.
And let's talk about these interactions then.
So chatbots obviously famous at presenting text that appears in many cases to be a human.
And I'm guessing that must be central to this.
Yes, so some of the ideas as to why this might happen are because of this.
This is a very human-like interaction that you're doing.
But also, these chatbots can be quite sycophantic.
So they can basically reinforce the things that you're already thinking.
And if your thinking is somewhat disordered already, if you're having paranoid delusions, the idea is that they could reinforce that.
So some researchers in the UK have proposed that conversations with chatbots can fall into a feedback loop.
So the AI reinforces the paranoid beliefs made by the users.
And in a preprint that they published in July, they simulated user chatbot conversations using prompts with varying levels of paranoia.
And they found that the users and the chatbots reinforced each other's paranoid beliefs.
Okay, and what are the companies who make these chatbots saying about this then?
Obviously, as you say, there are a lot of questions that remain to be asked and answered.
There are a lot of remaining questions for this and how exactly this might work.
The AI companies themselves have been looking into this.
And so, for example OpenAI, the company behind ChatGPT, They released a model on the 25th of April which was more sycophantic, and there were some reports then of cases of psychosis, and they rolled back that update, although it is still available for paid users.
They have announced though, that they're working on an update that would make ChatGPT de-escalate conversations that are not grounded in reality, and they've hired a clinical psychiatrist to help study the effects of the company's products on users' mental health.
Other companies, such as Character Technologies, which are behind Character AI, which is another one of these chatbots available.
They are including more safety features, resources for self-harm and features specifically for the safety of minors.
And Anthropic.
The company behind the chatbot Claude, has given the chatbot the ability to stop conversations if users resist the model's attempts to redirect conversation away from harmful or distressing topics.
Well, often in these chats, we say, well, this is something to keep an eye on.
More research is needed.
But I think, clearly in this case, this is one of those situations which we will continue to cover here at Nature.
And speaking of Nature, I read a paper on naturecom slash news, which is based on a paper posted to the preprint server bio archive last week.
Now, this, of course, means that this isn't peer reviewed research.
So something to bear in mind.
And this is another virus-related story for me this week, but very, very different.
This is about what researchers are saying is the first ever viruses designed by artificial intelligence.
Now, these viruses are capable of hunting down and killing strains of E. coli bacteria.
Wow, okay.
So when you say AI designed, is this someone going to chat GPT and say hey, design me a virus, or is it a bit more complicated than that?
A little bit more complicated than that.
You'll be shocked to hear, Nick.
But yeah, AI models have been used a lot to create DNA sequences, proteins and complexes with things interacting.
But a whole genome.
Now that's quite a lot more challenging because there are interactions between genes and gene replication and regulation processes.
Now, what the team have done here is they've used two AI models, EVO1 and EVO2, that analyze and generate DNA, RNA and protein sequences.
And they've designed genomes using a template right, a starting sequence that guides the AI to generate a genome with desired characteristics.
And in this case, they've used PhiX174.
Now, this is a virus, a bacteriophage or a bacteriophage, depending on how you pronounce it that infects E coli bacteria.
It's pretty simple, relatively.
A single-stranded DNA virus that contains just over 5300 nucleotides and 11 genes and all the genetic elements required to infect bacteria and replicate.
Little sidebar here.
This is one of the first genome sequences that was ever completed back in the 1970s.
I think it was the first complete DNA sequence.
And it was done in part by Fred Sanger, the double Nobel winner.
It took years.
Now you could do it in 10 minutes.
But anyway, this is their kind of template sequence that they've used to train the AI.
And so this is obviously a relatively simple thing for the AI to design.
But why is it that they were interested in getting an AI to design this rather than, you know, trying to do it themselves?
I guess?
I think the idea is that it could do a lot of stuff relatively quickly.
So what happened was?
The EVO models were trained on more than 2 million phage genomes, with some help from the researchers, using a method called supervised learning.
And what this spat out was PhiX174-like viral genomes that had specific functions that they were looking for, namely infecting these E coli bacteria.
And they evaluated thousands of these sequences then and narrowed their search down to 302 viable bacteriophages.
And these shared more than 40 nucleotide identity.
So DNA identity with PhiX174, but some had completely different coding sequences similar, but in some cases very different, i guess.
And so the researchers synthesized these dna sequences and inserted them into a host bacteria, which then meant the viruses could grow and replicate and so on.
And then they were experimentally tested to see whether they could kill e coli bacteria, and were they successful?
Could they kill e coli?
Some of them yes.
16 of the 302 AI-designed bacteriophages showed specificity for E coli and could infect the bacteria.
And the researchers showed that combinations of AI-designed phages could infect and kill three different E coli strains, which the wild-type Phi-X174 this was all kind of based on originally was unable to do.
And you know, E coli is kind of a workhorse of the lab, but it can also be kind of a bit of a horrible disease.
Was the idea that they wanted to make something that could help get rid of E coli, or was it just to see if they could do this?
Yeah, I think a little bit of both.
I think the authors say that this shows the potential for designing biochemical tools and therapies for treating bacterial infections, certainly in the face of things like antibiotic resistance.
Now, phages have been used in some cases to kill nasty bacteria, so i think that's one particular idea here.
But it has to be said in this case the e-coli that were being studied were non-pathogenic and are kind of regular lab workhorse, And researchers not involved in this work said that this is a good example of sort of what could be done today, where the state of AI is today, and maybe give an idea of what could be done in the future.
In their paper.
In their manuscript, the authors say that it lays the foundation for the generative design of useful living systems at the genomic scale, which I think is a ways off.
Obviously, this is a very simple thing this virus, still complicated in terms of all these moving parts, work together.
But designing anything more complicated will take time.
And the system itself isn't able to just work independently.
As I say, it needs intervention and guidance from the team.
But it does show that this could be a route for this to be done further in the future.
And so there are all sorts of rules and conventions in place for people experimenting with genetically modified organisms and that sort of thing.
Will there be similar sort of ethical discussions about AI-designed viruses and things like that?
Absolutely right.
I think there will.
This is one of the things that need to be considered because, of course, systems like this could potentially be used to design viruses that can harm humans.
This is called the dual-use dilemma.
I think it has to be said, though, that this isn't unique to AI research.
It's something that always needs to be considered with a lot of biological research.
But given this is a new technology, these debates will have to be had.
And also there are questions about biosafety, of course.
Now the researchers do talk about this in their manuscript and they say that they excluded any viruses that infect eukaryotic cells, which are cells like URI, for example, a workhorse of this kind of study and has quote a long history of safe use in molecular biology research, as they say in their study.
Well, if there's more debates going on between researchers on how to use this technology, i'm sure we'll be talking about it in the future, but i think that's all we've got time for.
Thank you ben, and listeners for more on those stories and for a link of where you can sign up to get more like them direct to your inbox.
Check out the show notes for some links And that's all for this week.
If you've enjoyed this podcast, why not let us know?
You can leave a review or a comment on your podcast app of choice, or you can send a message on X or Blue Sky.
We're at Nature Podcast, or send us an email to podcast at nature.com.
I'm Benjamin Thompson.
And I'm Nick Petrichow.
Thanks for listening.
This moment, Alex's mile swim, it means everything to him and to all of us.
At Evernorth Specialty Services, we help providers make sure patient care doesn't get interrupted, especially for rare, chronic and complex conditions.
Now Alex and his care team have his hemophilia under control with life-saving medication and care management from Evernorth's specialty pharmacy Acredo.
So Alex can be the last one out of the pool.
Evernorth Specialty Services, because every moment counts.
Visit evernorth.com slash specialty to learn more.
Hey, it's Maya and Sim from the Girls That Invest podcast.
If you are an IT or security pro, you know managing devices, identities and applications can feel overwhelming and honestly risky.
That's where Trellica by 1Password comes in.
It helps conquer SaaS brawl and shadow IT by discovering every app your team uses, managed or not.
With pre-built app profiles, you can assess risk, manage access and even optimize your software spend.
Plus, it simplifies onboarding, offboarding and compliance, all while cutting costs by eliminating unused licenses.
Take the first step to better security for your team.
Learn more at 1password.com slash special offer.
