FlashFeed

Full text

Researchers at the University of Toronto have demonstrated a computer worm that reasons its way across a network, working out a different attack for each machine it lands on, with no humans involved.

Much of the coverage has called this a breakthrough in demonstrating how the AI worm can target any online device, but that capability has been visible on the horizon for anyone watching the criminal underground.

What the paper does is settle the argument about whether it is buildable and, in doing so, reopens three issues most internet security teams would rather not confront directly.

The first of these is cost, and this is the story that matters. Building attacks aimed at specific systems used to be slow and expensive, requiring skilled operators.

That expense is one reason many mid-sized organizations were largely ignored - they were not worth the engineering effort. The Toronto worm removes that constraint by running an open-weight model on the GPUs of machines it has already compromised.

Devices too weak to host a model send their reasoning upstream to an infected node elsewhere on the network. The attacker’s compute bill is paid by the victims, and each captured machine extends the worm’s own infrastructure .

Once tailored attacks cost almost nothing, being uninteresting stops protecting you. Being reachable starts to matter far more than being interesting.

Patching complications

The next issue, patch management , is where things become a little more complicated. Traditionally, when a conventional worm would ride a specific vulnerability, the conventional response would be to patch it, and the worm would die. WannaCry spread across more than 150 countries in 2017 on a single flaw, reinforcing the lesson that rapid patching limits damage.

This worm, however, does not give defenders a single flaw to patch. It reasons out a different route per host, and in one experiment, when copies repeatedly failed on older systems because of a detection bug, the parent process identified the failing check, removed it, and tried again. You cannot rely on closing a single door against something that rewrites its approach as it goes.

And finally, and more worryingly, the worm was able to consume newly published security advisories during execution and generate attacks against vulnerabilities that did not exist when the underlying model was trained. That challenges the assumption that knowledge cutoffs meaningfully constrain offensive AI.

If the model can read today’s advisory and reason from it, the training date matters far less than many assume.

Working, with limitations

However, its limitations deserve to be stated plainly. Exploitation attempts succeeded 44 percent of the time, and the researchers noted that most failures were malformed payloads rather than flawed reasoning.

It was also slow. Yet according to the researchers, across fifteen experiments the worm obtained elevated access on about 74 percent of hosts, replicated onto roughly 62 percent, and reached seven generations of self-replication within a week.

A 44 percent success rate that continually retries is not a wall the threat runs into. It is a baseline, and that baseline tracks what open-weight models can do today, a capability that has so far moved in one direction.

Another underappreciated point is that the model runs inside an environment the attacker controls. That makes many of the safety mechanisms discussed by AI vendors (refusals, filters, and rate limits), largely irrelevant. There is nothing to rate-limit when inference is occurring on infrastructure the attacker already owns.

If your AI risk model assumes a platform provider will enforce guardrails, this is a scenario that will bypass them entirely.

Plan around the attacker

What to do about it is less dramatic than the threat itself. Keep patching. The problem is that patching was already a treadmill against capable adversaries, and this only makes the treadmill faster.

The more effective solution is to plan around an attacker that eventually gets in and adapts once inside. Segmentation and containment, for instance, should take priority over chasing individual vulnerabilities. If the entry point can be almost anything, then the critical question becomes how far an intrusion can spread.

There is also the issue of exposure. A worm like this will seek the easiest foothold first, just as human operators do. These can include leaked credentials, forgotten infrastructure, internet-facing services, or access already circulating in criminal markets.

Monitoring external exposure is primarily about buying time, rather than preventing compromise entirely, which is precisely what an autonomous attacker is designed to take away.

The worm remained in the lab, and the code is gated for defensive researchers, which was the right decision. But academic containment is not quite the same as reassurance.

The economics that once shielded many organizations are weakening, the patch-it-and-move-on model is losing effectiveness, and some researchers and practitioners are placing operationally relevant autonomous attacks within roughly twelve to eighteen months.

Therefore, the question we should be asking is whether to take this seriously now, on your terms, or later, on an attacker's.

We've reviewed and ranked the best endpoint protection software suites .

This article was produced as part of TechRadar Pro Perspectives , our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

Full text

There’s a reason why some of the most effective exercises tend to mirror movements in real life. It’s not because personal trainers and coaches lack imagination, but because the body doesn't care how creative your programming is — it cares whether you can climb a flight of stairs without grabbing the banister, for example, or if you can catch yourself from a stumble.

These are just a few of the benchmarks that matter in later life, and for elite performance coach David Higgins — who has trained everyone from Margot Robbie and Scarlett Johansson to Samuel L. Jackson, David Harbour, Game of Thrones' Richard Madden and the entire cast of The Batman , among many others — one exercise sits at the top of the list for anyone over 50: the step-up. Here’s why.

Lower-body power matters so much after 50

Most people understand that strength declines with age. What they underestimate, however, is how quickly it begins to matter in practical terms, despite not being able to be measured on the best smartwatches ' fitness age metrics.

“Lower body power becomes critical after 50 because we naturally lose muscle mass, reaction speed and balance as we age,” says Higgins. “The glutes, quads and calves are what keep you upright, stable and independent.”

If you can't generate force through the floor, he says, “everyday movements like climbing stairs, getting out of a chair or catching yourself from a fall become harder.” The step-up addresses all of this in a single movement: “it trains strength, balance and coordination all at once and mirrors real life better than most gym exercises.” It's slightly similar to the farmer's carry , an application of a real-life movement.

It’s also the reason Higgins places it above more popular alternatives. Walking is excellent, of course, but doesn't load the body enough to preserve muscle. Squats are bilateral — they share the work equally between both legs, which means they don't expose or address the kind of side-to-side imbalances that tend to develop quietly over decades. Step-ups, however, are unilateral, as each leg works independently, building the stability and gait mechanics that bilateral training misses.

“If I could only choose one lower-body movement after 50,” says Higgins, “step-ups would be near the top because they combine strength, stability, balance, gait mechanics and unilateral control in one movement; they bridge the gap between rehab and performance.”

Watch the tutorial video below for advice on how to do a step up:

The mistake most older adults make

Before thinking about adding height, load or additional reps, Higgins is more concerned with something more fundamental: the quality of the movement itself. “The biggest mistake older adults make is chasing fatigue instead of quality movement,” he says. “Rushing through reps, leaning through the hips, or pushing through the stronger side of the body.” Your nervous system, he says, “has to trust the movement before your body can own it.”

Rushing through step-ups doesn't just reduce the benefit — it reinforces the compensations you're trying to correct. If you're relying on the trailing leg to push you up, or leaning forward as you step, you're not actually building the unilateral strength and hip control the exercise is there to develop.

Instead, Higgins’ coaching cue is worth memorising: “push the floor away and finish tall through the glutes.” Most people pull themselves up through the knee, instead of driving through the hip. It's a subtle distinction, but makes an enormous difference to where the effort is actually made.

How to get started — and when to progress

(Image credit: Getty Images / Ildar Abulkhanov) To perform the step-up, stand facing a sturdy bench, box, or step. Place one foot fully on the platform, then drive through that heel to lift your body up until both feet are on the step. Step back down with control and repeat for the desired reps before switching legs. Keep your chest up and avoid pushing off excessively with the trailing leg.

For step height, beginners should start with a low box, roughly ankle to mid-shin height. "You want control, not compensation," he says. Fitter adults with solid mobility, balance and hip control can work up to knee height.

As for volume, Higgins prescribes 2–3 sets of 8–12 controlled reps per leg. "Tempo matters more than volume," he says. Drive through the whole foot, stand tall at the top and lower slowly under control.

Higgins confirms that just your bodyweight is sufficient resistance to begin with and, for many people over 50, it's exactly where they should stay — probably for longer than they'd expect. “Most people over 50 need to relearn movement patterns before adding load,” he says. “Once control, balance and posture improve, adding dumbbells or a weighted vest is a brilliant progression.” The dumbbells can wait. As for frequency? Two to three sessions a week is enough to see real benefit, Higgins says. “Consistency beats intensity every time.”

Precautions

Step-ups are accessible to most people, but Higgins flags a few situations worth considering. Anyone with severe knee osteoarthritis, significant balance issues or acute hip pain should approach the exercise carefully and ideally with professional guidance. “Often it's not that the exercise is wrong,” he says, “it's that the height is too ambitious or the body isn't controlling the movement properly yet.” Low and slow is always the right answer.

What leg-day moves are your go-tos in the gym? Have you tried the step-up yet? Let us know below.

Full text

Tom Holland thinks his Spider-Man movies can learn from The Odyssey by "finding ways to make people believe what they're watching is real"

"With every movie you make, you learn lessons," Holland, who plays Odysseus' son Telemachus in the adaptation of the Greek epic, tells GamesRadar+ at a press day for The Odyssey in London. "You find ways of working that are new that you might like, and [some] you don't."

Nolan's propensity for practicality, including utilizing the latest in IMAX technology, building a real Trojan Horse , and instructing his actors to sail , clearly rubbed off on Holland, who is set to star in his fourth MCU Spider-Man movie, Brand New Day , later in July.

"[Christopher Nolan]'s desire to be practical and to keep things in camera creates an atmosphere on-set which really inspires people to be creative and think outside the box," Holland says. "There are things I saw on this movie which I've never seen before and they're very simple. It just requires forward thinking and planning."

Holland, one of the veterans of superhero cinema with over half-a-dozen movies under his spandex belt, even admitted that his time globetrotting on The Odyssey with some of the business will shape his thinking the next time he plays Spidey – and he aims to pay it forward.

The actor remarks, "I come from the kind of superhero space, which I absolutely adore. I love playing [Spider-Man]. I think there are things that we could do in the future that can improve those movies by being more practical and finding ways to make people believe that what they're watching is real."

To that end, Holland reveals the ways Spider-Man: Brand New Day was influenced by his team on The Odyssey: "We've made a real suit, and we did real stunts, stuff like that."

Holland concludes, "I think that lots of people can learn from what Chris is able to do."

The Odyssey opens in cinemas on July 17.

For more, check out more of 2026's big movie release dates and our beginner's guide to The Odyssey . Then relive some of Nolan's classics with our ranking of the best Christopher Nolan movies .