AMD’s next-generation Ryzen AI 500 “Medusa Point” APUs are getting another step closer to reality, and this time Linux is doing the leaking. As first noted by Phoronix, AMD has published the firmware needed for its upcoming RDNA 4m integrated graphics, giving us another indication that the Zen 6-based mobile chips are nearing their debut.

RDNA 4m is a rather interesting hybrid

The new firmware targets AMD’s GFX 11.7 graphics IP, known as RDNA 4m. It sits in an unusual position between the existing RDNA 3.5 architecture and full-blown RDNA 4. AMD’s desktop RDNA 4 GPUs use the newer GFX 12 family, while RDNA 4m remains part of the GFX 11 branch.

The “m” is widely believed to stand for mobile, with RDNA 4m seemingly designed specifically for APUs. Earlier Linux and LLVM work suggests it brings some RDNA 4-style features, including FP8/BF8 conversion and WMMA instructions, potentially giving Medusa Point newer AI and graphics capabilities without a complete graphics architecture overhaul.

The new firmware supports GFX 11.7.0 and 11.7.1, along with security, display, and memory-management components, suggesting AMD is getting much of the software stack ready for the new chips. That’s especially interesting given previous leaks of a 10-core Zen 6 Medusa Point APU in Geekbench, which showed promising performance gains over current Ryzen mobile chips. AMD has confirmed the Medusa codename for its next-gen Zen 6 client platform, but the exact specifications and launch details are still under wraps.

The APU refresh is getting spicy

The biggest question now is what AMD can actually squeeze out of this RDNA 4m design. It isn’t a full RDNA 4 GPU, but the added instruction support could give AMD’s integrated graphics a useful feature boost while keeping the APU relatively efficient.

For now, the firmware landing ahead of launch is another promising sign that Medusa Point is moving from roadmap territory toward real hardware. If AMD can pair Zen 6’s CPU improvements with a more capable iGPU and modern AI features, its next Ryzen generation could make for some seriously interesting laptops.

Few of the top AI labs have published or demonstrated containment response plans, according to a recent study. A containment plan spells out what happens once an AI is caught trying to subvert human control — what access gets cut, and when the system gets shut down entirely.

That’s the finding from Guidelight AI Standards, an organization dedicated to promoting safe frontier AI development practices, which graded five leading labs on how prepared they are for exactly this scenario. OpenAI came out on top; Anthropic and Meta scored lowest. The findings matters as agentic AI takes on more autonomous roles inside companies’ own systems, and as regulators in California and New York begin requiring disclosure. For anyone building on or investing in these models, it’s a rare independent read on how seriously each lab treats operational risk versus how it talks about it.

Guidelight’s assessment was based on publicly available plans from Anthropic, Google, OpenAI, Meta, and xAI, graded across a range of metrics, including how well each company logs and monitors what its AI systems are doing internally, whether it halts systems after a surge of flagged misbehavior, whether independent third parties audit its controls and publish findings, and what its exact plan is for containing a model that goes off the rails.

Concern over whether AI companies can contain their increasingly capable and agentic models has grown in the wake of a series of high-profile cybersecurity incidents in which models from OpenAI, Anthropic, and Meta gained unintended access to the internet during safety evaluations and hacked into external systems. 

The findings highlight differences in how AI companies are publicly approaching safety as they scale up agentic deployment into environments where AI systems can take serious actions at scale. While some AI companies have detailed how they test their models for dangerous capabilities before deployment, they’ve generally been less vocal about what happens when models already operating inside their systems misbehave. 

“I was surprised by how little the AI companies have said about how they would handle a very serious incident if their model did escape their control in some sense,” Steven Adler, Guidelight’s chief scientist and former OpenAI safety researcher, told TechCrunch. 

Guidelight defines a containment plan as a “pre-specified plan, triggered when the AI is detected trying to subvert control, which covers what permissions to revoke from the model, who the model may continue operating for, under what constraints, and when to take it fully offline.”

“There’s good reason to think that the leading models at the frontier AI companies right now are misaligned in some sense,” Adler said. “Whenever the models are doing work on the company’s behalf, the company should have some scaffolding around it to be able to tell what that AI is doing, look for signs of misalignment, stop it from doing something very dangerous before it takes that action, and generally plan for what they would do in the event of a serious control incident where they have an emergency on their hands and need to figure out how to contain that loss of control incident.”

To date, most of the plans in place for managing catastrophic risk are still largely left up to the companies. Guidelight’s report says the best public evidence shows that companies have “few containment protocols ready for an emergency.” 

There could, of course, be containment plans that companies have in place but haven’t shared publicly. A Google spokesperson told TechCrunch the Guidelight report doesn’t represent the full scope of the company’s AI safety and security measures. The company did not respond to TechCrunch’s question of whether Google has an internal containment response plan that has not been publicly disclosed.

An OpenAI spokesperson mirrored similar sentiments, saying Guidelight’s assessment doesn’t capture all of the company’s internal practices. “We have a process for requiring restricting permissions, pausing workloads, limiting deployment, or taking the model fully offline, and have applied it,” the spokesperson said.

Meta declined to say whether it has an internal containment response plan, instead pointing TechCrunch towards an existing AI framework that outlines thresholds of risk and how it tests for loss of containment.

Lily Li, a privacy and AI lawyer and founder of Metaverse Law, told TechCrunch she believes companies might be hesitant to disclose the full scope of their containment policies and assessments on public-facing websites for legal, not just competitive, reasons.

“The concern from a company perspective is that if you make the disclosures too specific, and you’re not living up to your promises, that could form the basis of an unfair and deceptive marketing claim and expose you to more liability going forward,” Li said.

The point of Guidelight’s study is largely to encourage companies to be more transparent about their safety plans. Regulators are starting to force the issue, too.

California’s SB 53, which took effect this year, requires large frontier developers to publish frameworks explaining how they identify and respond to critical safety incidents and manage risks from models circumventing oversight mechanisms. New York’s RAISE Act, which has similar criteria, takes effect in January. 

Last month, representatives introduced the AI Kill Switch Act, a bipartisan federal bill that would require major AI developers to build and maintain technical mechanisms to shut down rogue AI models. 

“A kill switch is the bare minimum for today’s models,” said Connor Leahy, U.S. executive director of nonprofit ControlAI. “If the last few weeks revealed anything, it is that these companies don’t understand the systems they are building, and the models are growing to a point where they’re harder to rein in when they go rogue. Without a way to turn off the current dangerous systems, and with all the incentives to continue building more uncontrollable systems, we are heading in a very dangerous direction.” 

Without a containment plan in place, Adler said, companies might be figuring out their responses to an emergency on the fly and “winging it in response to this much faster adversary.”

Guidelight’s assessment measured whether each company implements six priority practices from its Control standard, based only on publicly available information — so a low score reflects a lack of public disclosure, not necessarily a lack of internal safeguards.

The companies with the lowest scores for publishing their containment plan were Meta and Anthropic — the latter perhaps more surprising than the former given Anthropic’s rhetoric on safety. Guidelight says Anthropic’s August Risk Report doesn’t mention “limiting the deployment of one of its models as one of the possible results of its process to investigate and respond to misalignment and control incidents.” Similarly, Guidelight was able to find no evidence that Meta has a containment response plan or has any plans to adopt one. 

An Anthropic spokesperson said that if the company detected a model attempting to evade oversight or otherwise subvert human control, it would conduct a risk assessment focused on determining whether containment is the appropriate response.

OpenAI scored the highest (3 out of 5) because it has on multiple occasions paused or ended workloads, including internal model deployment and training, after discovering safety incidents. It has also described what steps it would take before resuming workloads. 

“However, we have found no evidence that [OpenAI] has adopted a formal plan for when and how to respond to misalignment incidents in the future,” the report reads. 

Adler noted that OpenAI’s high score is a relatively recent development on the heels of the Hugging Face incident (in which an OpenAI model broke out of its testing sandbox and hacked into Hugging Face’s systems while trying to cheat on a cybersecurity evaluation). After that, the company shared more details about how it has cordoned off some of its misbehaving models. 

That episode is just one example of AI systems acting against the goals of the company that built them. Consider a separate case involving Anthropic’s models, which essentially tried to talk the maintainers of an open source codebase into accepting code with vulnerabilities.

Adler said such a circumstance could easily happen within an AI company’s internal systems. To prevent that, he suggests companies scan their AI system’s chain of thought — the model’s step-by-step reasoning — to look out for signs of deception, long-running plotting, or plans to introduce vulnerabilities into code that they can take advantage of later. 

The methods Guidelight is advocating for are very straightforward to implement, Adler says, and in many cases, versions of them already exist. “It’s about making the decision inside of the company to care enough about this risk to slightly broaden the scope,” Adler said. 

One of the main challenges is that researchers want to be able to operate flexibly within their AI systems, and introducing real-time, preventative monitoring could create friction. “Researchers basically do their thing, and if there’s an issue, someone else gets to clean it up afterward, and the researchers don’t have to change their workflow in the meantime,” he said.

The problem with “clean-up monitoring after the fact” is that it leads to researchers scrambling around to fix problems. And for some types of incidents, it might be too late. For example, an AI could turn off a company’s control system, which means researchers can no longer count on catching the misbehavior later. 

Many in the AI industry will complain that creating set plans to handle misbehavior is fundamentally difficult because AI moves too fast; today’s plans will be worthless tomorrow. 

Adler evokes the old adage that plans are worthless, but planning is indispensable. 

“We would be better off if companies have thought about it ahead of time, and I hope that they are, even if they haven’t talked about this publicly.”

xAI did not respond in time to comment.

Gav Free of the Slow Mo Guys has spent years chasing moments too fast for ordinary eyes. Several years back a hammer strike on a single mirror left him staring at the footage in confusion. Right at the moment of impact a small ripple spread across the surface. He kept returning to those frames, wondering how something so rigid could flex like that. This time he arrived with a Phantom TMX 7510, a pile of spare mirrors, and the determination to settle the question for good.

To shield him from flying shards, protective gear was placed up over him, including a flame suit donated by Adam Savage. There were lights on both sides of the camera, as well as another running parallel to a large sheet of glass against which cracks would be clearly evident. Later, a slew of Colored Titan tubes were used to paint the reflections in varying hues. The mirrors were mounted on cheap platforms, and the camera simply waited at frame rates of up to a fraction of a second, resulting in extended, mesmerizing sequences.

Insta360 X6, 8K 360 Camera with 1/1.1” Sensors

• DUAL 1/1.1″ SENSORS & TRIPLE AI CHIP SYSTEM: Custom-built dual 1/1.1″ square sensors deliver unprecedented 360 image quality. Combined with a triple…
• FLAGSHIP 8K50FPS 360 VIDEO: Hit record on our 360 action camera and capture your world in stunning native 8K50fps video. Every view, every highlight…
• IN-HOUSE AI MODEL FOR EFFORTLESS 360 EDITING: Insta360’s own AI model, built specifically for 360 editing. New AI Director edits your footage…

First strikes were happening at 120,000 frames per second. When the hammer hits the glass, a large wave comes out. The center pieces begin to spin away from the place of impact, and the hammer’s reflection becomes larger as the shards tip. Once the surface has bent far enough, which is pretty significant, the glass will break at the impact zone. The initial crack that forms is essentially a perfect circle and only extends as far as the wave has rolled. Beyond that line, the glass breaks in more random, jagged patterns. Even at this speed, the cracks nearly disappear in a few frames, making it difficult to notice their progress.

It’s not simply the speed, but also how forcefully or softly you hit the glass, which shapes the story, literally. A particularly forceful strike causes a larger ripple, which occasionally reaches the edge of the mirror first. When this happens, the initial cracks appear along the rim rather than in the center. A softer hit allows the wave to die out sooner and keeps the break more central. At 80,000 frames per second, two cracks that appear to be simultaneous to any conventional camera are actually four frames apart, with a 50 microsecond delay. It’s little, but the high-speed footage makes it clear.

Pushing the frame rate even higher, up to 875,000 frames per second at lower resolution, transforms the entire scene into something virtually abstract. Black and white frames with millisecond exposure periods sharpen the leading edge of each fracture. These fronts are moving in nearly perfect circles, with each point along the edge traveling at the speed of sound through glass. The circular shape holds because all directions advance at the same rate until the entire surface gives way.

When you look at it from the side, you notice the most surprising behavior. The hammer does more than just bounce off or punch through; it sinks into the surface for an extended period of time while the glass continues to bend. Looking through the mirror to a grid behind it clarifies the duration of contact even further. The surface is curved in smooth lines, rising and sinking like the skin of a pond after a stone has been dropped in. Only after that deep flex do cracks appear and begin to spread outward.

Colored lights bounce off the glass, suddenly transforming those moments into a surreal landscape. The stable blue and green tubes appear to contradict the frenzied dance of the other colors as they flicker, painting the waves and eventually shattering in an ever-changing tapestry of shifting patterns. The fragments swirl and catch the light, giving the entire collapse the appearance of an explosion of frozen color rather than broken glass. One scene in particular made Gav think of a glass peacock falling apart or a hammer crashing through someone’s most vivid dream.

Later frames demonstrate how the final crack pattern captures the mirror’s three-dimensional shape at the exact moment it gave way. Long, open cracks develop where the surface dips into a bowl, whereas tight groups of fine cracks appear where it rises into a hump. The shatter process appears to petrify the vibration pattern in place, much like sand on a vibrating plate settles into distinct, defined lines. Looking down from the top, you can nearly make out the height map of the glass by observing how the cracks had organized themselves.

$29 off a device built to end handwritten meeting notes for good is one of the more useful discounts around right now.

The Plaud Note Pro has dropped to $159.99, a 15% cut from its $189.00 list price, putting one of the most capable AI voice recorders on the market within easy reach of anyone tired of scribbling in meetings.