BackgroundAccurately predicting seizures remains challenging. With advances in smart medical technology, EEG-based monitoring has become essential. This study aims to improve prediction accuracy using a hybrid framework that models multiscale EEG characteristics.MethodsEEG signals are decomposed into multiple sub-bands using the Discrete Wavelet Transform, and representative time-frequency and nonlinear features are extracted. These features are fed into a channel-centric model integrating depthwise separable convolution, self-attention, and an enhanced bidirectional long short-term memory network (DC-SA-EBiLSTM). The architecture integrates depthwise separable convolution for local spatial feature extraction, multi-head self-attention for global inter-channel dependencies, and an enhanced BiLSTM for channel-wise sequence modeling. The proposed method was evaluated on the CHB-MIT dataset using a 10-fold cross-validation protocol. An event-level leave-one-seizure-event-out validation was also conducted to assess alarm-based prediction performance.ResultsThe proposed approach achieved an average accuracy of 95.89%, sensitivity of 96.70%, specificity of 95.48%, and AUC of 99.02%. In the event-level validation, the model achieved an event sensitivity of 95.96%, an average false alarm rate of 0.316 FPR/h, and a mean early warning time of 30.52 min.ConclusionThe DC-SA-EBiLSTM framework effectively captures local and global inter-channel dependencies and provides a feature-driven approach for patient-specific preictal state prediction, showing potential for EEG-based seizure prediction.
Personality functioning in adolescents with depression: links with childhood maltreatment, psychopathology, self-harm and suicidal ideation
BackgroundAdolescence represents a critical developmental period marked by significant vulnerability to depression, a condition with heterogeneous presentations that complicate clinical management. The co-occurrence of personality dysfunction with depression is known to indicate greater severity and poorer prognosis, yet specific features of this comorbidity in adolescent clinical samples require further delineation. This study aimed to compare clinical and psychosocial correlates in depressed adolescents with and without impaired personality functioning.MethodsThe clinical sample comprised 73 adolescents (aged 12–17; 83.6% female) diagnosed with depression or reporting clinically significant depressive symptoms. Participants completed self-report measures assessing personality functioning (LoPF-Q 12–18), childhood maltreatment (CEQ), psychopathology (YSR 11-18), mentalizing capacity (RFQY-8), borderline traits (BPFS-C), and self-harm behavior. Based on LoPF-Q 12-18 T-scores, participants were categorized into two subgroups: depression without personality dysfunction (T ≤ 64; n=20) and depression with personality dysfunction (T ≥ 65; n=53).ResultsResults indicated that adolescents with co-occurring depression and personality dysfunction exhibited significantly lower functioning across all personality domains compared to those with depression alone. This subgroup also reported an earlier onset and higher frequency of self-harm, more severe suicidal ideation, elevated borderline traits, and greater impairments in mentalizing. Furthermore, they demonstrated higher levels of internalizing, affective, conduct, and PTSD symptoms, alongside greater exposure to emotional neglect.ConclusionImpaired personality functioning in depressed adolescents is associated with elevated and multifaceted patterns of symptoms, including subjective distress and risky behaviors. These findings underscore the necessity of routinely assessing personality functioning in adolescents with clinically significant depressive symptoms to enable early identification and the development of tailored, integrated interventions for this high-risk subgroup.
Additive impulsivity and emotion dysregulation in adolescents with comorbid bipolar and substance use disorder: a cross-sectional factorial study
IntroductionComorbid bipolar disorder (BD) and substance use disorder (SUD) in adolescence is associated with poor clinical outcomes, yet the independent and interactive contributions of impulsivity and emotion dysregulation remain poorly understood.MethodsThis cross-sectional study employed a 2 × 2 factorial design to examine impulsivity, measured with the Barratt Impulsiveness Scale-11, and emotion regulation difficulties, measured with the Difficulties in Emotion Regulation Scale, across four groups of adolescents (N = 128; aged 12–18 years): BD+SUD (n = 32), BD-only (n = 32), SUD-only (n = 32), and healthy controls (n = 32). All clinical participants were assessed during euthymia. Factorial analyses of covariance controlled for age, sex, residence, family structure, and income.ResultsSignificant BD × SUD interactions were found for emotion regulation, F(1,120) = 35.89, p < .001, ηp2 = .230, and impulsivity, F(1,120) = 9.51, p = .002, ηp2 = .073. The BD+SUD group showed the highest scores on both measures, exceeding the SUD-only group by 38.90 points on emotion dysregulation and 26.72 points on impulsivity. In the substance-using subsample (n = 64), impulsivity was the strongest predictor of substance use severity (B = 0.61, p < .001; R2 = .48). The BD+SUD group also displayed earlier illness onset, mixed-feature predominance, greater polydrug use, and exclusive high-lethality suicide attempts. Low income was the strongest exploratory predictor of clinical group membership.DiscussionThese findings support an additive comorbidity model in which BD and SUD jointly amplify impulsivity and emotion dysregulation, and they highlight the need for integrated, impulsivity-focused interventions in adolescents with dual diagnoses.
Waiting between war and loss: a qualitative study on the experience of ambiguous loss among Syrian refugees
IntroductionAmbiguous loss refers to individuals’ lack of knowledge about the whereabouts or fate of loved ones due to traumatic events such as war or natural disasters. It is associated with significant economic, social, and psychological consequences, including grief, depression, and identity confusion.MethodsWe examined the psychological impact of ambiguous loss among 15 refugees displaced by the Syrian civil war. Participants (22 -53 years old; 9 women, 6 men) had experienced the disappearance or prolonged uncertainty regarding the fate of their spouse, child, sibling, or relative and had lived with uncertainty regarding the fate of their loved one for 4 to 13 years (in some cases, the uncertainty resolved after the interviews). We collected the data through semi-structured interviews and analyzed them using inductive thematic analysis.ResultsFindings revealed emotional oscillation between hope and hopelessness, social isolation, and difficulties adapting to changing roles. Although beliefs and collective rituals served as coping resources, inadequate support and stigma intensified distress.DiscussionThe results indicate that ambiguous loss is embedded in sociocultural contexts and underscore the need for culturally sensitive, family- and community-based psychosocial interventions and supportive national policies.
Family risk factors, dyadic coping, and family resilience in young stroke dyads: an actor-partner interdependence mediation model
ObjectiveStroke in young individuals constitutes a significant familial stressor, with ensuing family risk factors critically influencing recovery. While couples’ collaborative coping strategies play a vital role in enhancing family resilience, the mechanisms underlying this process remain largely unexplored. This study aims to elucidate the impact of familial risk factors and dyadic coping on the family resilience of young stroke.MethodsThis cross-sectional study integrates evidence-based research with clinical data. We developed a structured questionnaire grounded in identified familial risk factors. A cohort of 243 dyads, comprising young stroke patients and their spouses, was recruited from the neurology departments of four tertiary hospitals. The Actor–Partner Interdependence Mediation Model (APIMeM) framework was utilized, in which patients were treated as actors (influencing their own outcomes) and spouses as partners (influencing each other’s outcomes), to determine both direct and indirect effects among variables, with significance evaluated using bootstrap procedures (5,000 resamples, 95% confidence intervals). Model simplification was guided by chi-square difference testing.ResultsThe analysis indicated that family resilience could be strengthened in young stroke survivors and their spouses. APIMeM analyses showed that patients’ and spouses’ anxiety (β = –0.479, P = 0.002) and depression (β = –0.718, P < 0.001) had significant negative actor effects on their own family resilience, while patients’ activities of daily living (ADL) had significant positive actor (β = 0.175, P < 0.001) and partner effects (β = 0.198, P < 0.001) on family resilience. Crucially, dyadic coping partially mediated these relationships, with significant actor-actor indirect effects for anxiety (β =–0.139, 95% CI [–0.261, –0.029]), depression (β = –0.190, 95% CI [–0.321, –0.084]), and ADL (β = 0.062, 95% CI [0.009, 0.131]) on family resilience. Significant actor-partner mediation effects were also observed for depression (β = –0.072, 95% CI [–0.170, –0.017]) and ADL (β = 0.028, 95% CI [0.002, 0.079]).ConclusionsUsing a dyadic APIMeM approach, this study demonstrates that dyadic coping serves as a key mediator linking familial risk factors to family resilience in young stroke dyads. Resilience emerges as a shared, co-constructed process rather than an individual attribute, underscoring the need for interventions targeting the couple as a unit to mitigate intervention bias.
The Meta hack shows there’s more to AI security than Mythos
On June 5, 404 Media reported that attackers had been using Meta’s AI customer support agent to steal Instagram accounts. Their approach was simple: They asked the agent to link the accounts to email addresses that they controlled, and the agent complied. One attacker broke into the dormant Obama White House account and made pro-Iran posts; others took over accounts with valuable, single-word handles, possibly in order to sell them.
AI cybersecurity concerns are nothing new. Since Anthropic announced in April that its Mythos model was too good at hacking to be released to the general public, commentators, researchers, and federal officials alike have fixated on the idea that superpowered AI systems could lay waste to our computer infrastructure. That’s not quite what this Instagram hack was: There, AI was the target rather than the attacker, and the method was far simpler than anything Mythos would cook up. But as companies offload more work to AI, these comparatively unsophisticated attacks could wreak their own havoc.
“As AI becomes more and more widely used—especially when AI is more and more widely used to automate our work flows, like account recovery—I think attackers are going to be more and more motivated to attack AI itself,” says Neil Gong, a professor of electrical and computer engineering at Duke University.
Gong and other scholars have been issuing warnings about the security vulnerabilities of AI agents for a while. They publish papers and blog posts detailing exploits such as indirect prompt injection, which involves hijacking agents using commands hidden in websites, emails, or other seemingly anodyne data sources. Compared with these techniques, the Meta hack was practically mindless. The only complication that hackers had to overcome was using a VPN that matched the true account owner’s location; then they directly asked the support agent to change the account’s email address, and it complied.
Meta has not commented publicly on how this vulnerability slipped through the cracks. But given the simplicity of the exploit, Gong says, it should have been uncovered easily, before the agent was deployed. “It’s really surprising,” he says. “I don’t understand why they didn’t find this simple problem.”
Jessica Ji, a senior research analyst at Georgetown’s Center for Security and Emerging Technology, agrees. “It raises questions like: Were there even guardrails in place?” she says. “Did anyone think to test for this kind of scenario?” She notes that the oversight is particularly striking coming from a company like Meta, which has extensive expertise in both AI and cybersecurity. Meta did not respond to a request for comment for this article, but on Monday a Meta spokesperson said on X that the vulnerability had been resolved.
As embarrassing a moment as this might be for Meta in particular, it also highlights some core vulnerabilities shared by all AI agents. Unlike traditional software, agents can respond in flexible—and unexpected—ways to new circumstances, which is why they might be able to substitute for human customer support agents. But AI agents can also be tricked in ways that humans wouldn’t be, and because they can take real-world actions, those mistakes have consequences. “A human would say, ‘Okay, why do you want to change the email address?’ and maybe respond with a security question,” says Somesh Jha, a professor of computer science at the University of Wisconsin–Madison. “What is going on with these agents is they’re very eager to finish the task. It’s almost like some elementary school student who just wants to please the teacher.”
There are ways to mitigate the risks. Companies can use traditional software to build guardrails that make sure agents follow strict rules, such as always asking for answers to security questions before sending sensitive account information to a new email address. And the experts consulted for this article all agree that agents should undergo rigorous red-teaming, a process in which developers try their best to attack a system in order to discover its vulnerabilities before it is deployed.
But there are also countervailing forces. Companies want to deploy capable agents, and the more power an agent has—and the fewer guardrails it is subject to—the more work it can potentially take on. “Security and utility always have a trade-off,” says Bo Li, a professor of computer science at the University of Illinois Urbana-Champaign. And adequate red-teaming can be expensive. Defenders have to expend more resources than attackers do, because attackers only need to discover a single exploit, while defenders try to discover and patch as many as they can. When attackers are working toward something as valuable as a single-word Instagram handle, they’ll pour resources into finding exploits, so defenders have to spend even more money to protect that prize.
As AI models continue to improve, hardening their defenses might actually get easier. Though the probabilistic nature of large language models means that LLM agents will always be vulnerable to some forms of attack, a more sophisticated model might have identified an attempt to change the email associated with the Obama White House account as suspicious. And AI systems can be used for agent red-teaming, much as participants in Anthropic’s Project Glasswing use Mythos to identify vulnerabilities in their software.
Still, experts expect that the problem of securing AI agents will only become more pressing in the future. As agents grow more capable, companies that adopt them may want to give them more power, both to provide more services with fewer humans and to avoid being left behind by their competitors. In the fast-moving world of AI, the time needed to carefully secure risky agentic systems might seem like an unconscionable delay.
“Everybody wants to be the first to do something and just push things out without careful scrutiny and red-teaming,” Jha says. “I think it’s a very dangerous thing.”
Are AI chatbots making us lose control of our brains?
This week I’ve been at SXSW London. There’s been music, film, and a lot—and I mean a lot—of talk about AI. I also had the opportunity to sit down with Gloria Mark, a psychologist at the University of California, Irvine, who has spent the last 30 years studying how people interact with digital technologies.
Early in her career, the biggest concerns were the potential impacts of internet and email use on our brains. We may laugh those concerns off today, but it’s true that as the technologies became more ubiquitous and ingrained in our daily lives, our attention spans began to shrink.
Mark is worried that things are only getting worse. The title of our session was “Have we lost control of our brains?” Unfortunately, Mark told me, the answer is yes.
Around two decades ago, Mark started wondering about how our use of devices might affect our attention spans. She set up what she calls “living laboratories,” using sensors and trackers to monitor adult volunteers’ attention, mood, and behavior when they were using devices.
In 2003, she found that the average user had an attention span of around two and a half minutes. That’s how long people could spend focused on one thing before moving on to something else. “That surprised me at the time,” she told me during our session on Wednesday. “I thought: Wow, this is really short.”
But when she repeated the experiment in 2012, she found that attention spans had shrunk—all the way down to around 75 seconds on average, she said. In research she conducted between 2014 and 2020, attention spans shrank further still—to a mere 47 seconds, on average. Yikes.
And it’s not good for us. Mark told me that she’s found switching our attention so frequently is stressful. “We would have people wear heart rate monitors, and … we would see direct correlation between switching attention fast and stress going up,” she told me.
All this distraction makes it harder for us to get stuff done, too. “It just takes longer to do any single task if you’re switching your attention,” she told me. “It’s not great for performance. It’s not great for our emotional well-being.”
And that’s for adults. What about the effects of digital technologies on children? A few months ago, Meta (which owns Facebook and Instagram) and Google’s YouTube were ordered to pay millions of dollars in damages to a 20-year-old woman who had accused the companies of creating products that led her to develop a childhood addiction.
Just a couple of weeks ago, Meta settled another lawsuit, this one brought by a rural school district in Kentucky. The district had also accused the company of designing addictive products that were harmful to students and had sought more than $60 million to cover the costs of their mental-health needs. Around 1,200 other school districts are taking similar legal action against social media companies.
But social media isn’t all bad, all the time. It can provide opportunities for some people, including those from marginalized groups, to form connections that might otherwise be difficult. A 2024 survey of LGBTQ+ teenagers found that while some described social media as a place of rejection and fear, others described it as a place where they felt a sense of belonging, where they could develop friendships and cultivate their identity.
In truth, we can’t definitively say what effects using social media is having on children across the board, says Mark. “There have been lots and lots of studies, and the evidence is to date inconclusive,” she told me. (Despite what you might read in best-selling books on the subject.)
Mark is hopeful that large, long-term studies might finally start shedding a bit more light on this question. An effort of this nature is underway in Australia, which enacted a social media ban for under-16s at the end of last year.
Given this uncertainty over a 20-year-old technology, I wondered if Mark had any thoughts on the potential impacts of AI—an obviously much newer offering that within the space of a couple of years appears to have become deeply integrated into our digital lives.
She told me she’s worried.
When we put in effort to do something—such as evaluating or summarizing content—we’re doing what’s known as “depth of processing,” she told me. “When you’re actively engaged with information, you’re processing it on a very deep level,” she said. “Then you’re more likely to learn it, to understand it, [and] to retain it.”
That’s not happening when most people use AI bots like ChatGPT, Claude, and Gemini. When we ask these tools to write, summarize, or evaluate for us, we’re no longer doing that depth of processing. “You’re deferring your cognitive work to AI,” she said. “And it’s not good for us.”
The risk is that our cognitive abilities will weaken over time. “If you’re not constantly exercising your muscles, they can atrophy,” Mark said. “And that’s exactly what can happen with our minds.” People with weaker critical thinking skills are more likely to fall prey to misinformation, she added.
Interactions with AI-powered “synthetic companions” can be just as harmful. Relationships between human beings take work—time, effort, and understanding. None of that is needed if you’re forming a relationship with a sycophantic bot. The “muscle” we risk atrophying here is emotional intelligence, which surveys suggest is already on the decline, said Mark.
She’s not painting a particularly rosy picture.
“If we continue on this trajectory, attention spans are diminished, loneliness is rising, boredom is rising, emotional intelligence decreasing, and actually our sense of purpose, according to studies, is also decreasing,” she said.
Luckily, she thinks we can course-correct by changing our relationship with these technologies. The key factor is effort.
The more effort we put into something, the deeper the satisfaction we stand to gain, Mark told me. That means making an effort to read a book rather than skimming its summary, and to meet with friends in person when you can. Try not to use GPS in places where you can probably manage without it.
“I love technology; we can’t give it up,” she told me. “[But] we have to learn how to create new life routines.”
This article first appeared in The Checkup, MIT Technology Review’s weekly biotech newsletter. To receive it in your inbox every Thursday, and read articles like this first, sign up here.
Tiny HHS office tasked with protecting research participants’ safety is running on fumes
In 2021, the federal office charged with ensuring that the vast research enterprise bankrolled by the Department of Health and Human Services keeps study participants safe, received a report of a death by suicide involving a person enrolled in a study testing a treatment for depression and reduced mobility.
Once the Office of Human Research Protections started looking into the death at the New York State Psychiatric Institute, it found problems that spread far beyond that one study, including failures in how the institute’s ethics board reviews proposed research. “We saw, holy smokes, this whole institution has issues,” said Lisa Buchanan, then director of the OHRP’s compliance division. The investigation became all-consuming. “It was probably 60-70% of our time,” she said.
A Comparative Study of Driving Performance Among Adolescents With Autism Spectrum Disorder and Neurotypical Adolescents
Conditions: Autism Spectrum Disorders
Interventions: Other: Assessment of executive functions and attention; Other: Evaluation on a driving simulator; Other: Evaluation on-road
Sponsors: Hopital La Musse; French Road Safety Delegation
Not yet recruiting
Interventions: Other: Assessment of executive functions and attention; Other: Evaluation on a driving simulator; Other: Evaluation on-road
Sponsors: Hopital La Musse; French Road Safety Delegation
Not yet recruiting
CanSOLVE 2.0 Theme 3.1 Mind the Gap – Dialectical Behavioural Therapy (DBT) Skills Training Program Pilot
Conditions: Mental Health; Mental Health Care; Hemodialysis; DIALYSIS SYMPTOMS AND ANXIETY; Health Related Quality of Life
Interventions: Behavioral: Dialectical Behavioural Therapy Skills Training
Sponsors: University of Manitoba; Canadian Institutes of Health Research (CIHR)
Recruiting
Interventions: Behavioral: Dialectical Behavioural Therapy Skills Training
Sponsors: University of Manitoba; Canadian Institutes of Health Research (CIHR)
Recruiting