RSAC 2025: Cisco and Meta put open-source AI at the heart of threat defense

With cyberattacks, which speeds up the speed of the machine, Open Source Large Language Models (LLS) Quick start and global cyberecurity leaders have become infrastructure, which allows you to develop and place faster, efficient protection than human analysts.

Open source LLMS has created a wide range of a wide range of a wide range to provide a faster, more adaptation, more adaptation and lower price, lower price infrastructure. Last week RSAC 2025 Conference, Cisco, Meta and Projectiscasticy The new open source declared an invasion of an opening surface in a community and a society, which determines the future of the open source of cyber.

One of the Key Roads from RSAC this year is the turn of open source LLMs to expand and strengthen the infrastructure.

Open source AI, many cybersecurity leaders have been summoned for years, which is the years with the years of being able to join forces against increasingly complex threats. To be an employee in creating a combined, open source LLM and infrastructure, is one step closer to the ads in RSAC.

Ciscon’s chief crop worker Jeetu Patel, “The real enemy is not our opponent. It is really our enemy.

The pathel said the relevance of accepting a complex problem: “The EU changes everything in the heart of this. It does not deal with human-scale threats; these attacks occur in the car scale.”

Cisco’s Foundation-SEC-8B LLM identifies a new era of Open Source AI

Ciscon’s newly established Photo’s Ai Group stems from the company’s final receipt Firm intelligence. The Fund EU’s focus is to provide a special AI infrastructure specifically to cyberecurity applications that are the most difficult to solve the most difficult. The methane is based on the architecture of Llama 3.1, this is not an 8 billion parameter, a common large language model is not a general purpose. This is carefully designed, performed in a database related to a cybersecretality of a cyber-based cyber in Cisco Foundation AI.

“According to nature, these problems in this charter are some of the most difficult ones in the AI ​​today. writes Yaron singer, VP EU and security in the fund.

Open source anchored fund AI, Cisco has developed effective architectural approaches to cyberective providers who are usually compared to comparable solutions, which are usually compared to each other, create stronger defenses.

Singer writes“Regardless of your unique needs of the organization, regardless of your unique needs,” regardless of your unique needs of this. Cisco’s Blog Post The announcement of the model recommends security groups to apply the baseline-sec-8b for the security lives. Cisco recommends that potential usage recommends in Cisco, model SOC acceleration, active threatening protection, engineering security, AI auxiliary code reviews, configuration and special integration.

Foundation-SEC-8B weight and Tokenizer stems open under the Perisive Apache 2.0 license Hug faceThe seller who protects the configure and privacy control, which allows the enterprise’s level customization and placement, allows you to place without the lock. Cisco’s blog also plans to open a training pipelineThe community further enhances managed updates.

Cybersecurity is in LLM’s DNA

Cisco, SOC, Devsecops and large-scale security teams chose to create a special model for the needs of the teams. Available, returning the General AI model, it is possible to go to their goals, so the Foundation AI team has used a large-scale, wide-owned and well-designed cyberecurity.

By taking a more accurate approach to the installation of the model, the Fund AI team was able to ensure that the model realized the real world cyber dangers, weakness and defense strategies.

The basic training databases include:

• Weakness bases: Detailed crills (general weaknesses and expositions) and CWES (general weakness of weakness) to determine the well-known threats and weaknesses.
• Maps threatening behavior: Established from proven security frames such as Miter O and CKto provide context to aggressive methodologies and behaviors.
• Threatening intelligence reports: Global Kyberecurity’s events and comprehensive opinions arising from the emergence threats.
• Red-team playbooks: Tactical plans that reflect the world figure and penetration strategies in Real.
• Real-world event summaries: Documented analysis of cybersaluture violations, events and their reduction ways.
• Compliance and Security Rules: The National Standards and Technology Institute also established the best practices from the leading standard (Nist) Frames and Open World Application Safety Project (Oven) provide coding principles.

These appropriate training regime positions are superior in Foundations-SEC-8B, complex cybertance positions, expanded accuracy, deeper contextual understanding and faster threatening capabilities offer faster than general purpose alternatives.

Evaluation Foundation Cream-SEC-8B llm

Ciscon’s technical assessment Fund-SEC-8B provides cybersem performance comparable to larger models:

Calister	Foot-SEC-8B	Call-3.1-8b	Call-3.1-70B
Honorary	67.39	64.14	68.23
CTI-RCM	75.26	66.43	72.66

To relate to the stock model cyberCetity, CIBERSECO allows high infrastructure expenses to obtain Cisco, SOC teams, without paying more efficiency with advanced threatening analysts.

Details on its blog, a broader strategic vision of Cisco, Stock AI: Strong intelligence for cyberem, General AI integration problems, limited domain alignment of general purpose models, inefficient data and difficulties inherit system integration. FOOSS-SEC-8B is specially designed to walk these barriers that work effectively over minimal device configurations that usually require only one or two Nvidia A100 GPUS.

Meta also stressed the extensive source strategy of AI defenders to expand the AI ​​defenders, which expanded AI defenders in RSAC 2025. Their open source tool set includes now LlBut the protection of 4, AI Workstreams are a multimodal classifier that reveals policy violations between text and images, improving compliance monitoring.

Also the present LlamafiEewall, Open source, real-time security framework that integrates into the modular capabilities Submission 2used to reveal urgent injection and jailbreak attempts. Also started as part of the llamafirewall Agent Alignment Checks This monitor and AI agent protect the decision-making processes CodeshieldDesigned to check the code generated to identify and alleviate the sensitivity.

Meta has developed 2 developed proposals in the 2 open source options that further strengthen the future of the AI-based infrastructure. It is a high-precision 86m-parameter model and lean, low-delay, less, low delay, which has an optimized alternative for minimal resource use, low delay.

In addition, Meta launched the open source benchmarking set Cybersec Eval 4made in partnership with crowding. FEATURES CyberosuAI effectiveness in the Scenarios of the Real Security Operations Center (SOC) and Car parkThe software is used to assess the autonomous AI capabilities to determine and correct the weakness.

Meta, open-based security vehicles, including sensitive document classifier and sound threat, launched the program to the program. Personal Processing is a privacy in the EU’s first and device in WhatsApp.

In RSAC 2025, ProjectDiscovery won the “most innovative start” award in the Sand Box, stressed the commitment to open source Kyberecurity. Its flagship vehicle, NocLei, APIS is a customizable, open source weakness scanner, which is a global society that quickly identifies the vulnerabilities of websites, cloud environments and networks.

Nuclei’s Wide YAML-based Template Library More than 11,000 examples of detection, 3000 have a real-time danger identification, connecting directly to specific peaks. Andy Cao, COO stressed the strategic importance of open source in predictions, connected: “Open source models of the 20th annual annual RSAC innovation Sandbox can succeed in cyberspace. This reflects the power of the society to be managed in democratization.”

Matches ProjectDiscovery’s success 2024 Hype cycle for Gartner’s Open Source ProgramIn the “Innovation Trigger” stage, the open source of open sources of AI and cyberships. Gartner recommends organization’s open source software offices (OSPOS), program documents (SBOM) frames (SBOM) frames (SBOM) and ensure regulation compatibility through effective management practices.

Effective concepts for safety leaders

Cisco’s Basic-8B, Meta’s expanded AI defenders set and the cores on the proximated avenue, the openness of the cybersem innovation, cooperation and specialized domain experience, demonstrated the most developing the company’s borders. Others, such as companies and others, are a stage to actively cooperate in creating cyberecurity protection that provides more effectiveness in lower costs.

As the Patel stressed during his cheynu, “they do not dream.