MAROKO133 Eksklusif ai: OpenAI adds layered safeguards as frontier AI reaches higher cyber

📌 MAROKO133 Update ai: OpenAI adds layered safeguards as frontier AI reaches highe

AI is racing ahead in cyberspace, and defenders are scrambling to keep up.

OpenAI says its cybersecurity-focused models are rapidly advancing, with CTF performance jumping from 27 percent on GPT-5 in August 2025 to 76 percent on GPT-5.1-Codex-Max in November 2025.

The spike shows how quickly AI systems are acquiring technical proficiency in security tasks.

The company expects future models could reach “High” capability levels under its Preparedness Framework.

That means models powerful enough to develop working zero-day exploits or assist with sophisticated enterprise intrusions.

In anticipation, OpenAI says it is preparing safeguards as if every new model could reach that threshold, ensuring progress is paired with strong risk controls.

Defense-first strategy

OpenAI is expanding investments in models designed to support defensive workflows, from auditing code to patching vulnerabilities at scale.

The company says its aim is to give defenders an edge in a landscape where they are often “outnumbered and under-resourced.”

Because offensive and defensive cyber tasks rely on the same knowledge, OpenAI says it is adopting a defense-in-depth approach rather than depending on any single safeguard.

The company emphasizes shaping “how capabilities are accessed, guided, and applied” to ensure AI strengthens cybersecurity rather than lowering barriers to misuse.

OpenAI notes that this work is a long-term commitment, not a one-off safety effort. Its goal is to continually reinforce defensive capacity as models become more capable.

Layered security stack

At the foundation, OpenAI uses access controls, hardened infrastructure, egress restrictions, and comprehensive monitoring. These systems are supported by detection and response layers, plus internal threat intelligence programs.

Training also plays a critical role. OpenAI says it is teaching its frontier models “to refuse or safely respond to requests that would enable clear cyber abuse,” while staying helpful for legitimate defensive and educational needs.

Company-wide detection systems monitor for potential misuse. When activity appears unsafe, OpenAI may block outputs, redirect prompts to safer models, or escalate to enforcement teams.

Both automated tools and human reviewers contribute to these decisions, factoring in severity, legal requirements, and repeat behavior.

The company is also relying on end-to-end red teaming. External experts attempt to break every layer of defense, “just like a determined and well-resourced adversary,” helping identify weaknesses early.

Strengthening the ecosystem

OpenAI is building broader cybersecurity initiatives alongside internal safeguards. A trusted access program will soon allow qualified cyberdefense users to access enhanced model capabilities under controlled conditions.

The company is also testing Aardvark, an agentic security researcher that scans full codebases for vulnerabilities and suggests patches.

OpenAI says Aardvark has already uncovered novel CVEs and will offer free support to select nonprofit open-source projects.

To reinforce governance, OpenAI is forming a Frontier Risk Council, an advisory group of seasoned defenders who will help determine the boundary between responsible capability and misuse risk.

Through the Frontier Model Forum, OpenAI is working with other labs to build a shared threat model for frontier AI systems.

This effort aims to map how models could be weaponized, where bottlenecks exist, and how the industry can coordinate defenses.

Together, these initiatives reflect OpenAI’s long-term mission: ensuring the rising power of AI translates into real leverage for defenders—grounded in real-world needs, shaped by expert input, and deployed with care.”

🔗 Sumber: interestingengineering.com

📌 MAROKO133 Hot ai: MIT Researchers Unveil “SEAL”: A New Step Towards Self-Improvi

The concept of AI self-improvement has been a hot topic in recent research circles, with a flurry of papers emerging and prominent figures like OpenAI CEO Sam Altman weighing in on the future of self-evolving intelligent systems. Now, a new paper from MIT, titled “Self-Adapting Language Models,” introduces SEAL (Self-Adapting LLMs), a novel framework that allows large language models (LLMs) to update their own weights. This development is seen as another significant step towards the realization of truly self-evolving AI.

The research paper, published yesterday, has already ignited considerable discussion, including on Hacker News. SEAL proposes a method where an LLM can generate its own training data through “self-editing” and subsequently update its weights based on new inputs. Crucially, this self-editing process is learned via reinforcement learning, with the reward mechanism tied to the updated model’s downstream performance.

The timing of this paper is particularly notable given the recent surge in interest surrounding AI self-evolution. Earlier this month, several other research efforts garnered attention, including Sakana AI and the University of British Columbia’s “Darwin-Gödel Machine (DGM),” CMU’s “Self-Rewarding Training (SRT),” Shanghai Jiao Tong University’s “MM-UPT” framework for continuous self-improvement in multimodal large models, and the “UI-Genie” self-improvement framework from The Chinese University of Hong Kong in collaboration with vivo.

Adding to the buzz, OpenAI CEO Sam Altman recently shared his vision of a future with self-improving AI and robots in his blog post, “The Gentle Singularity.” He posited that while the initial millions of humanoid robots would need traditional manufacturing, they would then be able to “operate the entire supply chain to build more robots, which can in turn build more chip fabrication facilities, data centers, and so on.” This was quickly followed by a tweet from @VraserX, claiming an OpenAI insider revealed the company was already running recursively self-improving AI internally, a claim that sparked widespread debate about its veracity.

Regardless of the specifics of internal OpenAI developments, the MIT paper on SEAL provides concrete evidence of AI’s progression towards self-evolution.

Understanding SEAL: Self-Adapting Language Models

The core idea behind SEAL is to enable language models to improve themselves when encountering new data by generating their own synthetic data and optimizing their parameters through self-editing. The model’s training objective is to directly generate these self-edits (SEs) using data provided within the model’s context.

The generation of these self-edits is learned through reinforcement learning. The model is rewarded when the generated self-edits, once applied, lead to improved performance on the target task. Therefore, SEAL can be conceptualized as an algorithm with two nested loops: an outer reinforcement learning (RL) loop that optimizes the generation of self-edits, and an inner update loop that uses the generated self-edits to update the model via gradient descent.

This method can be viewed as an instance of meta-learning, where the focus is on how to generate effective self-edits in a meta-learning fashion.

A General Framework

SEAL operates on a single task instance (C,τ), where C is context information relevant to the task, and τ defines the downstream evaluation for assessing the model’s adaptation. For example, in a knowledge integration task, C might be a passage to be integrated into the model’s internal knowledge, and τ a set of questions about that passage.

Given C, the model generates a self-edit SE, which then updates its parameters through supervised fine-tuning: θ′←SFT(θ,SE). Reinforcement learning is used to optimize this self-edit generation: the model performs an action (generates SE), receives a reward r based on LMθ′’s performance on τ, and updates its policy to maximize the expected reward.

The researchers found that traditional online policy methods like GRPO and PPO led to unstable training. They ultimately opted for ReST^EM, a simpler, filtering-based behavioral cloning approach from a DeepMind paper. This method can be viewed as an Expectation-Maximization (EM) process, where the E-step samples candidate outputs from the current model policy, and the M-step reinforces only those samples that yield a positive reward through supervised fine-tuning.

The paper also notes that while the current implementation uses a single model to generate and learn from self-edits, these roles could be separated in a “teacher-student” setup.

Instantiating SEAL in Specific Domains

The MIT team instantiated SEAL in two specific domains: knowledge integration and few-shot learning.

• Knowledge Integration: The goal here is to effectively integrate information from articles into the model’s weights.
• Few-Shot Learning: This involves the model adapting to new tasks with very few examples.

Experimental Results

The experimental results for both few-shot learning and knowledge integration demonstrate the effectiveness of the SEAL framework.

In few-shot learning, using a Llama-3.2-1B-Instruct model, SEAL significantly improved adaptation success rates, achieving 72.5% compared to 20% for models using basic self-edits without RL training, and 0% without adaptation. While still below “Oracle TTT” (an idealized baseline), this indicates substantial progress.

For knowledge integration, using a larger Qwen2.5-7B model to integrate new facts from SQuAD articles, SEAL consistently outperformed baseline methods. Training with synthetically generated data from the base Qwen-2.5-7B model already showed notable improvements, and subsequent reinforcement learning further boosted performance. The accuracy also showed rapid improvement over external RL iterations, often surpassing setups using GPT-4.1 generated data within just two iterations.

Qualitative examples from the paper illustrate how reinforcement learning leads to the generation of more detailed self-edits, resulting in improved performance.

While promising, the researchers also acknowledge some limitations of the SEAL framework, including aspects related to catastrophic forgetting, computational overhead, and context-dependent evaluation. These are discussed in detail in the original paper.

Original Paper: https://arxiv.org/pdf/2506.10943

Project Site: https://jyopari.github.io/posts/seal

Github Repo: https://github.com/Continual-Intelligence/SEAL

The post MIT Researchers Unveil “SEAL”: A New Step Towards Self-Improving AI first appeared on Synced.

🔗 Sumber: syncedreview.com