OpenAI Unveils GPT-5.4-Cyber for Defensive Cybersecurity

OpenAI has unveiled GPT-5.4-Cyber, a variant of its latest flagship model fine-tuned specifically for defensive cybersecurity work. This follows rival Anthropic's announcement of its frontier artificial intelligence model Mythos a week prior.

Credit: UNSPLASH

Anthropic's Mythos is being deployed as part of "Project Glasswing," a controlled initiative. Under this program, select organisations are permitted to use the unreleased Claude Mythos Preview model for defensive cybersecurity purposes.

This Anthropic model has found "thousands" of major vulnerabilities in operating systems, web browsers, and other software. OpenAI, creator of popular chatbot ChatGPT, stated that GPT-5.4-Cyber will initially be rolled out on a limited basis.

It will be available to vetted security vendors, organisations, and researchers because of its more permissive design. The company is also expanding its Trusted Access for Cyber (TAC) program.

This expansion will reach thousands of verified individual defenders and hundreds of teams protecting critical software. OpenAI is adding new tiers to its TAC program, which was launched in Feb.

Higher levels of verification will unlock more powerful capabilities for users. Users approved for the highest tier will gain access to GPT-5.4-Cyber.

This model has fewer restrictions on sensitive cybersecurity tasks, such as vulnerability research and analysis.

---

• OpenAI unveiled GPT-5.4-Cyber, a model fine-tuned for defensive cybersecurity, which is being rolled out on a limited basis.

• The rollout is limited to vetted security vendors, organisations, and researchers.

• The company is expanding its Trusted Access for Cyber program with new tiers.

Source: REUTERS

Neuron-Powered Chips Learn Doom, Advance Biological Computing

Human brain cells grown on computer chips have successfully learned to play the classic first-person shooter game, Doom. This achievement brings biological computers closer to practical applications, such as controlling robotic arms.

Credit: Cortical Labs

Australian organisation Cortical Labs previously used neuron-powered chips to play Pong in 2021. Those chips consisted of clumps of more than 800,000 living brain cells grown on top of microelectrode arrays capable of both sending and receiving electrical signals. Researchers painstakingly trained the chips to control the game's paddles.

Cortical Labs has since developed an interface simplifying the programming of these chips using Python, a popular language. Independent developer Sean Cole then used Python to teach the chips to play Doom, accomplishing this in approximately one week.

Brett Kagan of Cortical Labs stated, "Unlike the Pong work that we did a few years ago, which represented years of painstaking scientific effort, this demonstration has been done in a matter of days by someone who previously had relatively little expertise working directly with biology." He added, "It’s this accessibility and this flexibility that makes it truly exciting."

The neuronal computer chip used about a quarter of the neurons compared to the Pong demonstration. While it played Doom better than a randomly firing player, its performance remained far below that of top human players.

The chip learned significantly faster than traditional, silicon-based machine learning systems and is expected to improve with newer learning algorithms, according to Kagan. He also noted that comparing these chips to human brains is not useful, explaining that the biological material processes information in ways silicon cannot replicate.

Andrew Adamatzky, from the University of the West of England in Bristol, UK, commented, "Doom is vastly more complex than earlier demonstrations, and successfully interacting with it highlights real advances in how living neural systems can be controlled and trained."

Steve Furber, from the University of Manchester, UK, agreed that Doom represents a significant step up from Pong. However, he highlighted that much remains unknown about how these neurons play the game, including how they understand expectations or "see" the screen without eyes.

Yoshikatsu Hayashi, from the University of Reading, UK, called the jump in capability exciting, bringing biological computers closer to useful real-world applications, such as controlling a robotic arm with biological computers, a task which Hayashi and his colleagues are attempting with a similar computer made from jelly-like hydrogel. Hayashi and his colleagues are attempting a similar task with a computer made from a jelly-like hydrogel. He stated, "[Playing Doom] is like a simpler version of controlling a whole arm."

Adamatzky further emphasised, "What’s exciting here is not just that a biological system can play Doom, but that it can cope with complexity, uncertainty, and real-time decision-making." He added that this is "much closer to the kinds of challenges future biological or hybrid computers will need to handle."

---

• Human brain cells on a chip successfully learned to play Doom.

• Cortical Labs developed an interface, allowing independent developer Sean Cole to program the chips using Python in approximately one week.

• The chip's performance surpassed random players and learned faster than silicon-based systems, despite not matching human players.

Source: NEW SCIENTIST

Roblox to Introduce Age-Based Accounts for Enhanced Child Safety

Videogame platform Roblox will create tailored accounts for young users, aiming to bolster online safety. This initiative follows intense global scrutiny regarding child endangerment concerns.

Credit: ROBLOX

Through its age-check system or parental verification, Roblox will assign users aged 5 to 8 to "Roblox Kids" accounts. Users aged 9 to 15 will be assigned to "Roblox Select" accounts.

The introduction of these age-based accounts is scheduled for early June. Chief Safety Officer Matt Kaufman stated, "We will also introduce, at the same time, new requirements on what content standards must be met in order to have content or games appear in either the Roblox Kids account or the Roblox Select account."

Roblox has faced significant criticism from governments worldwide over recent years. These accusations concern the organisation's alleged failure to protect children from sexual predators and exploitation.

Content appearing in "Roblox Kids" accounts will be limited based on maturity levels. Games must go through a "rigorous three-step review process" with strict eligibility requirements for creators, Kaufman said.

The chat feature for "Roblox Kids" accounts is turned off by default. For "Roblox Select" accounts, the chat feature is gradually turned on, depending on the user’s age.

Developers participating in the three-step review process must verify their IDs, enable two-step verification, and maintain an active Roblox Plus subscription.

Roblox announced on Friday that it will launch a subscription plan on April 30. This plan, which requires an active Roblox Plus subscription for developers, will cost USD 4.99 per month and offer discounts on in-game items, avatars, and additional platform benefits.

---

• Roblox will introduce "Roblox Kids" and "Roblox Select" accounts for users aged 5-8 and 9-15 respectively, starting in early June.

• The new accounts aim to enhance online safety following global scrutiny over child endangerment concerns.

• Content for "Roblox Kids" accounts will be maturity-limited and subject to a "rigorous three-step review process."

Source: REUTERS