RSAC 2025: CISCO and Meta Meta International Open Source Organization at the heart of defending the threat

With the acceleration of electronic attacks quickly, the machine, LLMS large language models (LLMS) Soon the infrastructure that enables startups and global cybersecurity leaders to develop and spread adaptive and effective defenses against threats that develop faster than human analysts can respond.

The initial LLMS open features of sources have created the fastest time in the market, a greater ability to adapt and a lower cost, mainly for developmental and safe to provide infrastructure. Last week Rsac 2025 conference, Ciscoand Dead and Projectdissovery A new open source LLMS company has announced the innovation of the surface of the attack by the community, which together determines the future of the open source in cybersecurity.

One of the main fast food of this year is the shift in open source LLMS to extensively extend and enhance infrastructure.

Artificial intelligence is openly open to the provision of what many cybersecurity leaders called for years, which is the ability of many cybersecurity providers to unite the forces against increasingly complex threats. The vision of being collaborators in the creation of a uniform LLM and an open source and the source is a closer step, given the ads in RSAC.

“The real enemy is not our competitor. It is actually a opponent. We want to make sure that we can provide all kinds of tools and to have the ecosystem division together so that we can combat the opponent collectively,” said Cisco Jetutu Patel.

Patel explained the urgency of facing such a complex challenge, saying: “Amnesty International mainly changes everything, and cybersecurity is at the heart of everything. We are no longer dealing with human scale threats; these attacks occur across the machine.”

Cisco’s Sec-LLM sets a new open source of artificial intelligence

Cisco is the newly established AI Group Foundation It arises from the recent acquisition of the company Strong intelligence. Foundation AI focuses on providing the AI’s infrastructure for the field designed explicitly for cybersecurity applications, which are among the most challenging. On the structure of the Meta’s Llama 3.1, this parameter is 8 billion, the open language of the weight of the weight is not an artificial intelligence for general purposes. He was designed for this purpose, he was carefully trained in a group of data for cybersecurity under the company’s sponsorship by Cisco Foundation AI.

“By its nature, the problems in this charter are one of the most difficult problems in artificial intelligence today. To make technology within reach, we decided that most of the works we do mainly should be open. Open innovation allows double effects throughout the industry, and plays an important role in particular in the field of cybersecurity,” He writes Yaron Singer, Vice President of Amnesty International and Security at the Foundation.

Through AI Open Source Installation, CISCO designed an effective architectural approach to cybersecurity providers who usually compete with each other, and sell similar solutions, to become collaborators in creating more unified defenses.

singer He writes“Whether you are including it in existing tools or building a completely new workflow, the foundation-SEC-8B adapts to the needs of your unique organization.” Cisco Blog post The announcement of the form recommends that the Foundation-CEC-8B safety teams apply via the security cycle. Possible cases of use include CISCO for an acceleration model, pre -emptive defense of the threat, engineering enabling, and software -backed code with artificial intelligence, compositions, and allocated integration.

SEC-8B and Takenizer weights are opened under the permissible APache 2.0 license EmbroideryAllow allocation and publication at the institution level without locking the seller, and maintaining compliance and privacy controls. The Cisco Blog also notes plans for open training tubesIncreasing the enhancement of innovation that society moves.

Cyber ​​security in LLM DNA

CISCO has chosen a model for cybersecurity, optimized for the needs of SOC, Devsecops and large -scale security teams. The update of the current artificial intelligence model will not get them their goal, and therefore the AI’s main team designed its training using a wide, expanded and spaian data set.

By following a more accurate focus on building the model, the AI’s foundation team managed to ensure that the model is deeply understood by electronic threats in the real world, weaknesses and defensive strategies.

The main training data sets included the following:

• Weak databases: Including detailed CVES (common weaknesses and exposure) and CWES (common weakness) to determine known threats and weakness.
• Devices of threat behavior: An organizer of the proven security frameworks such as Att & ckAnd providing a context on the methodology of the attacker and behaviors.
• Threat intelligence reports: Comprehensive visions derived from global cybersecurity and emerging threats.
• Red Books: Tactical plans that define rival techniques in the real world and penetration strategies.
• Incident summaries in the real world: Documented analyzes of cybersecurity, accidents and mitigation paths.
• Compliance and security instructions: The best practices created from the leading standards bodies, including the National Institute for Standards and Technology (Nest) The frameworks and safety project open all over the world (OwaspThe principles of safe coding.

The designer training system puts this Corporation-SEC-8B uniquely to excel in complex cybersecurity, which provides greatly enhanced accuracy, and an understanding of a deeper context and response capabilities to the fastest threat of alternatives to general purposes.

SEC-8B LLM Standards Corporation

Technical standards at Cisco Foundation-CEC-8B offer cybersecurity similar to much larger models:

standard	Basis-SEC-8B	Llama-3.1-8B	Lama -3.1-70B
Cti-mcqa	67.39	64.14	68.23
CTI-RCM	75.26	66.43	72.66

By designing the foundation model to be a cybersecurity, CISCO SOC teams enables greater efficiency with advanced threat analyzes without having to pay high infrastructure costs to obtain them.

Cisco’s broader strategic vision, detailed in its blog, AI: Strong intelligence for cybersecurity, It addresses the challenges of common integration of artificial intelligence, including aligning the limited field of models for general purposes, insufficient data groups and the difficulties of the integration of the old system. Foundation-SEC-8B is specially designed to navigate these barriers, and efficiently operate them on lower devices formations, usually requires one or two only Nafidia A100 graphics processing units.

Meta has also confirmed its open source strategy in RSA 2025, expanding the suite of artificial intelligence defenders to enhance security through artificial intelligence infrastructure. An open source tool group now includes totoAma Guard 4, A multi -media work that discovers political violations through text and images, which improves monitoring of compliance with the functioning of artificial intelligence.

It was also presented LamaWalls, An open source safety framework in the actual time merges the normative capabilities that include Orderguard 2Which is used to detect fast injections and prison attempts. It was also launched as part of Llamafirewall Checks alignment the agent Which monitors and protects decision -making operations AI, along with CodeshieldIt is designed to inspect the code created to identify and reduce security gaps.

Meta also strengthened the face guard 2, as it provided two open source variables, which increases the reinforcement of the future infrastructure based on the open source. It includes a high -resolution model of 86 m the parameter and an alternative less than 22 meters of improved parameters to use minimal resources.

In addition, Meta launched an open source measurement suite CYBERSEC EVAL 4And developed in partnership with Crowdstrike. He distinguishes CYBERSOC EVALAnd measuring the effectiveness of artificial intelligence in the scenarios of the Security Operations Center (SOC) AutopatchbenchWhich is used to assess independent artificial intelligence capabilities to determine and install weaknesses in programs.

Meta also launched the Llama Defenders program, which provides early access to open safety tools, including sensitive works of close materials and the sound threat detection. Special treatment is privacy first, artificial intelligence is tried on the device inside WhatsApp.

In RSA 2025, Projectdissovery won the “Most Innovated Emerging” award in the innovation box, highlighting his commitment to open source cybersecurity. Her pioneering performance, NoCfor me , It is a customizable and open source scanner driven by a global community that quickly defines weaknesses through application programming facades, web sites, cloud environments and networks.

Wide -based Nawa Library It includes more than 11,000 discovery patterns, 3000 directly associated with specific CVES, allowing the actual time threat to be determined. Andy Cao, Operations Manager at ProjectDistovery, stressed the strategic importance of the open source, male“The victory in the twentieth annual RSACCAC Sandbox sand is proven that open source models can succeed in cybersecurity. It reflects the strength of our approach that society drives to give the democratic character to security.”

The success of Projectdissovery is in line with NoiseWhich puts open source AI and cyberspace tools in the “Trigger Trigger” stage. Gartner recommends that institutions create open source programs (OSPOS), adopt the frameworks of the technical material for programs (SBOM), and ensure organizational compliance through effective governance practices.

Practical visions of security leaders

The Cisco’s Sec-Meta, Meta from Ai Defenders Suite and ProjectDismissovery Nuclei together that creating cybersecurity thrives more when openness, cooperation and specialized field experience across the company’s borders. These companies and others like them put the way for any cybersecurity provider to be an active collaborator in creating cybersecurity defenses that offer greater effectiveness at lower costs.

Patel stressed during the keynote, “These are not fantasies. These are realistic examples that will be delivered because we now have custom security models that will be affordable for everyone. A better security event will come at a cost of cost with the latest reasons.”